Preliminary Information

idt

C6
TM

Processor
Data Sheet
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

This is Version 1.10 of the IDT WinChip C6 processor datasheet.

The latest versions of this datasheet may be obtained from
www.winchip.com

© 1998 Centaur Technology, Inc. All Rights Reserved

Centaur Technology, Inc. reserves the right to make changes in its

products without notice in order to improve design or performance
characteristics.
This publication neither states nor implies any representations or
warranties of any kind, including but not limited to any implied warranty
of merchantability or fitness for a particular purpose. No license, express or
implied, to any intellectual property rights is granted by this document.
Centaur Technology, Inc. makes no representations or warranties with
respect to the accuracy or completeness of the contents of this publication
or the information contained herein, and reserves the right to make
changes at any time, without notice. Centaur Technology, Inc. disclaims
responsibility for any consequences resulting from the use of the
information included herein.

Trademarks
WinChip, IDT-C6, C6, and CentaurHauls are trademarks of Integrated Device Technology
Corporation.

AMD, the AMD logo, and combinations thereof are trademarks of Advanced Micro Devices, Inc.
Am486 is a registered trademark, and AMD-K6 is a trademark of Advanced Micro Devices, Inc.
Microsoft and Windows are registered trademarks of Microsoft Corporation.
Intel, the Intel logo, and combinations thereof are trademarks of the Intel Corporation. MMX and
Intel486 are trademarks of the Intel Corporation. Pentium is a registered trademark of the Intel
Corporation.
Cyrix, the Cyrix logo, and combinations thereof are trademarks of the Cyrix Corporation. Cyrix
6x86MX is a trademark of the Cyrix Corporation.
Other product names used in this publication are for identification purposes only and may be
trademarks of their respective companies.
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

REVISION HISTORY

Date Ver Item Revision

09/1997 1.0 Initial release
03/1998 1.1 Table 1-6 Updated max power dissipation
Figure 6-1 Changed marking to IDT WinChip C6
Figure 6-2 Added, “CPGA pinout (Top Side View)”
Table 6-1 Updated pins A12-A19
Figure 6-3 Removed dimension “A”
Section 7.1 Updated case temperature “Tc”
Section 7.3 Clarified procedure for applying thermocouple
Table 7-2 Revised Max Power
Section B-2 Added “Bus Data with Inactive Byte Enables”
Errata F-8 Added “FPU FRNDINT does not round 2^63 - 0.5 correctly”
Errata F-9 Added “FPU FPREM1 may get wrong sign”
Errata F-10 Added “FPU FIST,FISTP rounds certain numbers incorrectly”
Errata I-10 Added “BIST return code (EAX) may indicate false failures”
Errata I-11 Added “EFLAGS.TF incorrectly causes trap on entry to SMM”
Errata I-12 Added “RDTSC may return incoherent values in EDX:EAX”
Errata I-13 Added “RDTSC hangs when certain event counters are used”
Errata I-14 Added “EFLAGS.RF incorrectly cleared after external interrupt”
Errata I-15 Added “SMI# during REP string does not set EFLAGS.RF”
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

CONTENTS

1 INTRODUCTION ...................................................................1-1
1.1 Basic Features .................................................................1-1
1.2 Processor Versions ..........................................................1-2
1.3 Competitive Comparisons ................................................1-3
1.4 Compatibility.....................................................................1-6
1.5 Data Sheet Assumptions ..................................................1-7
2 ARCHITECTURE...................................................................2-1
2.1 Introduction.......................................................................2-1
2.2 Key Concepts ...................................................................2-2
2.3 Component Summary.......................................................2-3
2.3.1 General Architecture ..................................................2-3
2.3.2 I-Cache .......................................................................2-5
2.3.3 Translator Unit............................................................2-5
2.3.4 Execution Unit ............................................................2-6
2.3.5 D-Cache .....................................................................2-7
2.3.6 X86 Fetch Unit............................................................2-7
2.3.7 FP-Unit .......................................................................2-8
2.3.8 MMX-Unit....................................................................2-8
2.3.9 Bus-Unit......................................................................2-8
3 PROGRAMMING INTERFACE..............................................3-1
3.1 General.............................................................................3-1
3.2 Additional Functions.........................................................3-3
3.3 Machine-Specific Functions .............................................3-3
3.3.1 General.......................................................................3-3
3.3.2 CPUID Instruction.......................................................3-4
3.3.3 Extended CPUID Instruction Functions ......................3-7
3.3.4 EDX Value After Reset. ..............................................3-8
3.3.5 CR4 ............................................................................3-9
3.3.6 Machine-Specific Registers......................................3-10
3.4 Omitted Functions ..........................................................3-10
3.4.1 Pentium Appendix H Enhancements ........................3-11
3.4.2 Other Functions........................................................3-12

CONTENTS i
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

4 HARDWARE INTERFACE.....................................................4-1
4.1 Bus Interface ....................................................................4-1
4.1.1 Differences .................................................................4-1
4.1.2 Clarifications...............................................................4-3
4.1.3 Omissions...................................................................4-3
4.2 Signal Summary ...............................................................4-4
4.3 Power Management..........................................................4-7
4.3.1 Static Power Management..........................................4-7
4.3.2 Dynamic Power Management.....................................4-7
4.4 Test & Debug ...................................................................4-8
4.4.1 Machine Check...........................................................4-8
4.4.2 BIST............................................................................4-8
4.4.3 Internal Error Detection ..............................................4-9
4.4.4 JTAG ..........................................................................4-9
4.4.5 Debug Port ...............................................................4-10
5 ELECTRICAL SPECIFICATIONS..........................................5-1
5.1 AC Timing Tables for 75-MHz Bus ...................................5-1
5.2 AC Timing Tables for 66-MHz Bus ...................................5-3
5.3 AC Timing Tables for 60-MHz Bus ...................................5-6
5.4 DC Specifications .............................................................5-8
5.4.1 Recommended Operating Conditions ........................5-8
5.4.2 Maximum Ratings .......................................................5-9
5.4.3 DC Characteristics ...................................................5-10
5.4.4 Power Dissipation.....................................................5-10
6 MECHANICAL SPECIFICATIONS ........................................6-1
6.1 CPGA Package ................................................................6-1
7 THERMAL SPECIFICATIONS...............................................7-1
7.1 Introduction.......................................................................7-1
7.2 Typical Environments .......................................................7-1
7.3 Measuring TC ....................................................................7-1
7.4 Estimating TC ....................................................................7-2
7.5 Recommended Thermal Solutions ...................................7-4
7.6 Contacts ...........................................................................7-5

ii CONTENTS
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

APPENDIX A. MACHINE SPECIFIC REGISTERS ............................... A-1

A.1 General ................................................................................. A-1
A.2 Category 1 MSRs.................................................................. A-4
02h: TR1 (Pentium Processor Parity Reversal Register)...... A-4
0Eh: TR12 (Pentium Processor Feature Control) ................. A-4
10h: TSC (Time Stamp Counter) .......................................... A-5
11h: CESR (Control & Event Select Register) ...................... A-5
12h-13h: CTR0 & CTR1 (Event Counters 0 & 1) .................. A-6
107h: FCR (Feature Control Register).................................. A-6
108h: FCR2 (Feature Control Register 2)........................... A-10
109h: FCR3 (Feature Control Register 3)........................... A-10
A.3 Memory Configuration Registers ........................................ A-11
General ............................................................................... A-11
Memory Configuration Registers ........................................ A-11
MCR Control Register......................................................... A-14
APPENDIX B. COMPATIBILITY........................................................... B-1
B.1 Introduction ........................................................................... B-1
B.2 Bus Compatibility .................................................................. B-2
B.3 Integer Instruction Compatibility ........................................... B-4
B.4 Floating-Point Compatibility .................................................. B-5
APPENDIX C. PROCESSOR ERRATA ................................................ C-1
C.1 Introduction ........................................................................... C-1
C.2 Floating-Point Errata............................................................. C-4
C.3 MMX Errata......................................................................... C-11
C.4 Integer Errata...................................................................... C-13
C.5 Bus Errata........................................................................... C-22

CONTENTS iii
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

1 INTRODUCTION

The IDT WinChipTM C6TM processor, designed by Centaur

Technology Inc., is a plug-compatible alternative to the Intel
Pentium processor and to the new Intel Pentium processor
with MMX TM technology (also known informally as the P54 and
the P55 processors).
The IDT WinChip C6 processor family is based on a unique
Centaur-developed design approach and is manufactured in the
IDT 0.35-micron CMOS technology. This combination provides
high-performance, low-cost, and low-power solutions to the
desktop and mobile personal computer market.
In addition to the Intel Pentium family, the IDT WinChip C6
processor also directly competes with the AMD-K6TM and Cyrix
6x86MXTM processors. When considered individually, the
performance, cost, and power dissipation of the IDT WinChip C6
processor family are all very competitive. When considered as a
whole, the IDT WinChip C6 processor family offers a
breakthrough level of value.

1.1 BASIC FEATURES

The IDT WinChip C6 processor family comprises several
versions. All family versions share the following common
features:
n Plug-compatible with the Intel Pentium processor—bus,
electrical interface, and physical package (“socket 7”).
n Software-compatible with Intel Pentium processors and the
thousands of X86 software applications
Notable n Software-compatible with the new Intel MMX technology
Features n Two large (32-KB each) on-chip caches.
n Performance optimized for business applications (as
represented by the ZD Winstone97 Business benchmark)
n Lower power than the equivalent Pentium processor.
n Low cost due to a very small die (88 mm2) and low-cost
technology.

INTRODUCTION 1-1
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

1.2 PROCESSOR VERSIONS

The IDT WinChip C6 processor family consists of a single
functional version that is suitable for both desktop and mobile
applications. This basic version is offered in several internal
speed ranges and two different voltage settings.

n Plug-compatible with the Intel P54 processor in a Ceramic

PGA package—bus, electrical interface, and physical
Desktop package.
Features n Operates in a Socket 7 (the standard socket that accepts both
a P54 and a P55 processor).
n Like the P54, uses only one system-board voltage (versus the
P55 which requires two voltages).

n Same package and interface as for desktop

Mobile
Features n Provides lower power consumption than the corresponding
Mobile Pentium processor with MMX technology.

n The IDT WinChip C6 processor is available in four speed

grades:
Speed
180-MHz (60-MHz bus),
Versions
200-MHz (66-MHz bus)
225-MHz (75-MHz bus)
240-MHz (60-MHz bus)

n IDT WinChip C6 processors support one of two voltage

ranges:

Voltage 3.52V (3.45V–3.6V)

Versions 3.3V (3.135V–3.6V)

1-2 INTRODUCTION
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

1.3 COMPETITIVE COMPARISONS

The following tables summarize the major features of the IDT
WinChip C6 processor and its primary competitors. The
competitive information is as specified in the competitive
processor’s data sheets and is accurate only as of the time this
datasheet was written. The features are those that characterize
the primary capabilities of an x86 processor.
The major themes of this summary are:
n The IDT WinChip C6 processor has equivalent or better
cache and TLB capabilities. These are critical to system
performance for modern PC operating systems and
applications. (See Table 1-1)
n The IDT WinChip C6 processor has a very simple instruction
execution architecture. This results in a very small die size
for the functions provided (and thus facilitates its low price
and low power). (See Table 1-2)
n The IDT WinChip C6 processor has very competitive internal
frequency. This high frequency, combined with the large
caches, yields very good system performance. (See Table 1-3)
n The IDT WinChip C6 processor has a much smaller die size
in the same basic technology. This small size benefits the
user by facilitating low price and low power consumption.
(See Table 1-4)
n The IDT WinChip C6 processor has superior (reduced) power
dissipation. (See Table 1-5)

INTRODUCTION 1-3
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

Table 1-1. Cache and TLB Characteristics.

Major Features C6 P55 K6 M2
I-Cache Size 32 KB 16 KB 32 KB 256
Data Ways 2 4 2 assoc
D-Cache Size 32 KB 16 KB 32 KB 64 KB
unified
I & D cache
Data Ways 2 4 2 4
TLB Size (I / D) 64 / 64 32 / 64 64 /128 16
Ways 4 assoc ? direct
L2 TLB N N N 64 x 6
Page Dir Cache 8 entries N N N

Table 1-2. Microarchitecture Characteristics.

Major Features C6 P55 K6 M2
Decode Single 2 Insts 3 Insts 2 Insts
In-order In-order In-order In-order
Issue & Execute Single 2 Insts 3 insts 2 Insts
In-order In-order Out-order Out-order
Branch Prediction N Y Y Y
Call/Return Stack 8 entries Y 16 entries N

Table 1-3. Performance Characteristics

Major Features C6 P55 K6 M2
Clock Multiplier 2x, 3x, 4x 2x, 3x, 2x, 3x, 2x, 3x,
Ratios (vs. Bus 5x 3/2x, 5/2x 3/2x, 5/2x 3/2x, 5/2x
frequency)
Bus Speed (MHz)s 60,66, 60,66 66 66,75
75
Max Internal Freq 240 MHz 233 MHz 233 187.5
(233PR)

1-4 INTRODUCTION
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

Table 1-5. Technology

Major Features C6 P55 K6 M2
Technology Metal 0.35µ 0.35µ 0.35µ 0.35µ
Poly 0.28µ
4LM 4LM 5LM+LI 4LM
Voltage I/O 3.3/3.52 3.3 3.3 3.3
Core same 2.8/2.45 2.9/3.2 2.9
Die Size 88 mm2 128 mm2 162 mm2 197 mm2

Table 1-6. Power Dissipation

Major Features C6 P55 K6 M2
MHz 200 200 200 166
(PR200)
Worst Case W 15.7 20.0 20.2
3.52V 13 (2.8V) (2.9V) (2.9V)
3.3V 10.5

INTRODUCTION 1-5
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

1.4 COMPATIBILITY
The IDT WinChip C6 processor is compatible with both the Intel
Pentium processor and the new Intel Pentium processor with
MMX technology. However, the IDT WinChip C6 processor is
specifically optimized for desktop and mobile PC configurations
(as opposed to server environments).
An IDT WinChip C6 processor can plug into existing Pentium
processor-based desktop and portable system boards and can
operate without requiring changes to the system hardware. In
some cases, a special BIOS may be needed (due to possible use
by the BIOS of Pentium processor-unique machine specific
registers). Currently, BIOS support for the IDT WinChip C6
processor is available from Award, AMI, and SystemSoft.
The IDT WinChip C6 processor does not provide Pentium-
compatible multiprocessing (neither do the mobile Pentium
processor, the AMD-K6 processor, and the Cyrix 6x86MX
processors).
Note that all processors developed for use in PCs (“x86”
processors) have some differences in low-level functions. (These
include differences between the various Intel processors; and
between these processors and the equivalent Cyrix and the
AMD processors.) The IDT WinChip C6 processor has similar
differences.
Centaur has performed extensive testing of hundreds of PC
boards, peripherals, software applications, and operating
systems to confirm the IDT WinChip C6 processor’s
compatibility.
Indicative of this compatibility, the IDT WinChip C6 processor
has obtained the XXCAL Inc. Platinum Certification (their
highest compatibility rating) and Microsoft Windows 95
certification.

1-6 INTRODUCTION
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

1.5 DATA SHEET ASSUMPTIONS

The IDT WinChip C6 processor specifications are directly based
upon the Pentium processor’s external specifications as defined
by: (1) publicly available Intel publications, and (2) by the actual
behavior (derived from testing) of the Pentium processor. This
datasheet book provides only minimal descriptions of these
Pentium-compatible functions. The major emphasis in this
document is to describe differences from the explicit and implicit
(behavioral) Pentium specifications.
The intent of these specifications is to make it easy for a board
designer, system designer, or BIOS developer to utilize the IDT
WinChip C6 processor in place of the Pentium processor or the
Pentium processor with MMX technology. (This, of course,
makes it trivially easy for the end-user to be able to exploit the
advantages of the IDT WinChip C6 processor.) We assume that
the reader is a potential direct user of the IDT WinChip C6
processor and is thus familiar with the specifications of the
Pentium processor.
Table 1-7 lists some relevant documents that define the
reference x86 architecture.

Table 1-7. x86 Architecture Specification Documents

Document Title Intel Version
Order #
Intel Architecture Software Developer’s Manual, Vol. 1 243190 001
Intel Architecture Software Developer’s Manual, Vol. 2 243191 001
Intel Architecture Software Developer’s Manual, Vol. 3 243192 001
Pentium Processor Family Developer’s Manual 241428 005
Pentium Processor with MMX Technology 243185 004
Pentium Processor Specification Update 242480 027

INTRODUCTION 1-7
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

2 ARCHITECTURE

2.1 INTRODUCTION
Even though the IDT WinChip C6 processor is externally
compatible with the Pentium processor, the internal
architecture and design of the IDT WinChip C6 processor is very
different from that of the Pentium processor and other
contemporary x86 processors such as the AMD-K6 and Cyrix
6x86MX processors. The IDT WinChip C6 processor uses a
unique design approach that provides significant benefits to the
end-user.
This design approach provides high performance at low cost and
low power using a relatively simple architecture that runs at
high internal clock frequencies (MHz), includes large on-chip
caches, and is extensively optimized for the target PC
environment. The resulting IDT C6 processor is smaller (die size
is only 88 mm2 in 0.35µ geometry technology) and uses less
power than any contemporary X86 processor.
Philosophically, this is a return to the same basic concepts of
RISC design that allowed microprocessor performance
breakthroughs in the 1980’s. Recently, however, contemporary
x86 processors have followed a different path using very
complex internal designs employing advanced architecture
concepts such as superscalar execution, out-of-order instruction
execution, reorder buffers, non-blocking caches, and so forth
(these terms are all found in the datasheets of competitive
products).
Unfortunately, while these advanced technical concepts make
for good technical reading, the real bottom-line benefit that they
provide to the end-user has been limited; especially when
considering the resultant large chip sizes (resulting in high
costs) and high power consumption. No such advanced technical
hocus-pocus is to be found on an IDT WinChip C6 processor — it
merely offers compatibility with good performance, very low
costs, and very low power consumption.

ARCHITECTURE 2-1
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

2.2 KEY CONCEPTS

The key concepts underlying the IDT WinChip C6 processor
design are:
n Simple instructions (load, store, branch, ALU) dominate
instruction execution time. This is the basic RISC design
concept, which is also true in the x86 architecture: over 90%
of instructions executed come from these basic categories. Of
course, “simple” x86 instructions are more complex than
corresponding RISC architecture instructions.
The IDT WinChip C6 processor optimizes performance of
these types of basic x86 instructions while minimizing the
hardware provided for other little-used x86 functions. The
little-used instructions are primarily implemented in
microcode with minimal hardware support.
n Improving clock frequency has higher leverage than
improving CPI. The result of advanced computer design
approaches over the last few years has been that the
improvements in cycles-per-instruction (CPI) come at the
expense of cycle-time improvements such that total
performance (a linear function of frequency and CPI) of
complex designs (“brainiacs”) hasn’t improved as fast as
performance of simpler, yet high clock frequency, designs
(“speed demons”).
The IDT WinChip C6 processor optimizes total system
performance by generally optimizing for highest clock
frequency, even if this means reduced CPI for some functions
versus more complex processor designs.
n Memory performance is the limiting CPI performance factor.
Due to the high ratio of internal clock speed versus the
relatively limited PC processor-bus speed, off-chip memory-
access performance is the primary factor in processor CPI
performance (as opposed to internal instruction execution
performance).
The IDT WinChip C6 processor addresses this phenomenon
by providing very large on-chip caches that run at the high
internal processor clock frequency. In addition, sophisticated
TLB and cache management algorithms are included to
further reduce bus activity.

2-2 ARCHITECTURE
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

n Optimize the design for the target user environment. The IDT
WinChip C6 processor implements very specific and detailed
design tradeoffs to provide high performance with low cost.
Minimal hardware is provided for functions that are not
heavily used or that are not critical to performance in the
target environments (low-end desktop and mobile systems).
These design optimizations are based on extensive and
detailed analysis of the actual behavior of Windows
operating systems and applications
n Small is beautiful. The IDT WinChip C6 processor is highly
optimized for small physical size and few logic transistors. In
addition to the obvious cost benefits, this small size provides
secondary benefits of low power consumption and improved
reliability.

2.3 COMPONENT SUMMARY

2.3.1 General Architecture

Figure 2-1 illustrates the basic components of the IDT WinChip
C6 processor.
Fundamentally, the IDT WinChip C6 processor’s internal design
is a relatively simple five-stage pipeline execution core with an
additional instruction translation stage to translate x86
instructions coming from the fetch stage into the internal micro-
instruction format.
Fetching and translating x86 instructions is asynchronous to
the internal execution pipeline. Instructions are issued one at a
time in program order. Instructions are executed and retired in
order. Cache misses stall the pipeline until the data is available
for the requesting instruction.
In spite of this very basic micro-architecture, the IDT WinChip
C6 processor achieves very high performance through several
mechanisms:
n High internal clock frequency. The design is heavily
optimized to provide high frequency.
n Good CPI on highly used instructions. The IDT WinChip C6
processor implements specific design features to reduce the
number of cycles for heavily used instructions — including
complex functions such as protect-mode segment-register
loads and string instructions.
n Large and fast on-chip caches and TLBs.

ARCHITECTURE 2-3
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

n Lots of fine-tuning and low-level optimizations. This includes

such items as fast unaligned data access and fewer pipeline
interlocks than the Pentium processor.
Stage

RET Stk I-TLB I-Cache I

X86
Fetch

X86 Inst Buffer

X86-To-Microcode
Translator X

Queue

Trap X86 X86

Addr Inst Immed
R Parms Data
ROM
F
Bus Bus
Unit Internal Insts

Decode R
PDC

Address Gen A
Address
D-Cache FP MMX
Access Pipe Pipe D
& Execute
Writeback W

D-TLB D-Cache

BFRS

Figure 2-1. IDT WinChip C6 Processor Data Flow

2-4 ARCHITECTURE
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

2.3.2 I-Cache
The I-cache contains 32 KB organized as two-way set associative
with 32-byte lines. An LRU replacement algorithm is used. The
associated I-TLB contains 64 entries organized as a 4-way set
associative with a pseudo-LRU replacement algorithm. This
large cache has a one-clock access time and operates at the high
clock frequencies of the IDT WinChip C6 processor.
The I-TLB utilizes an eight-entry unified page directory cache
that significantly reduces the TLB miss penalty. In addition, the
I-cache control logic includes several innovative features that
minimize cache invalidates and unnecessary bus fetches.
As opposed to many other contemporary x86 processors, the data
in the I-cache is exactly what came from the bus; that is, there
are no “hidden” pre-decode bits. This facilitates the provision of
large cache capacity in a small physical size.
The I-cache is dynamically turned off when not used to reduce
power requirements.

2.3.3 Translator Unit

The I-cache or bus unit delivers 16 or 8 bytes per clock to an x86
instruction buffer in the translator unit. The translator converts
x86 instructions to internal instruction and data forms.
Assuming that the instruction is in the x86 instruction buffer at
the start of the cycle, the translator translates an entire x86
instruction in one clock. Instruction prefixes require an
additional translator cycle for each prefix. However, due to the
asynchronous fetch and “lookahead” capability of the translator,
these extra cycles for prefixes rarely result in a bubble in the
execution pipeline.
The output of the translator is: (1) the internal micro-instruction
stream to perform the x86 instruction function, (2) the
immediate data fields from the x86 instruction, and (3) various
x86 state information used to control execution (for example,
operand size). The internal instruction stream for an x86
instruction can consist of micro-instructions directly generated
by the translator, or micro-instructions from the on-chip ROM,
or both. For performance-sensitive instructions, there is no
delay due to access of micro-code from ROM.
The microcode ROM capacity is larger than most x86 microcode
ROMs to allow more unimportant (relative to performance)
functions to be performed in microcode (versus in hardware), to
allow extensive self-test microcode, and to allow extensive built-
in debugging aids (for processor design debug).

ARCHITECTURE 2-5
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

Instruction fetch and translator operation is made asynchronous

from micro-instruction execution via a three-entry translated-
instruction queue between the translator and the execution
unit. This queue allows the translator to “look-ahead” and
continue translating x86 instructions even though the execution
unit is stalled or is busy with a microcode sequence. The
translator can also overlap generation of multiple internal
instructions with translating prefixes on the subsequent
instruction.

2.3.4 Execution Unit

Internal micro-instructions are executed in a tightly coupled
four-stage pipeline that is very similar in structure to a basic
RISC pipeline:
n Decode stage (R): Micro-instructions are decoded, register
files are accessed, resource dependencies evaluated, and so
forth.
n Addressing stage (A): Memory addresses are calculated
and sent to the cache units. The IDT WinChip C6 processor
is capable of calculating most x86 instruction address forms
in one clock; forms containing two registers (or a shifted
index register) require two clocks.
n Execute stage (D): ALU operations or load accesses to the
D-cache are performed. All basic ALU functions take one
clock except multiply and divide.
n Write-back stage (W): The results of operations are
committed to the registers and store data is written to the D-
cache or external write buffers.
Although the pipeline structure is similar to non-x86 processors,
the micro-instructions and associated execution units are highly
tuned to the x86 architecture. The micro-instructions closely
resemble the corresponding x86 instructions. Examples of
specialized hardware features supporting the x86 architecture
are: hardware handling of the x86 condition code, segment
descriptor decode and manipulation instructions, hardware to
automatically save the x86 floating-point environment, and so
forth.

2-6 ARCHITECTURE
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

2.3.5 D-Cache
The D-cache is very similar to the I-cache: 32 KB organized as
two-way set associative with 32-byte lines. An LRU replacement
algorithm is used. The associated D-TLB contains 64 entries
organized as 4-way set associative with a pseudo-LRU
replacement algorithm. This large cache has a one-clock access
time and is designed to operate at the high clock frequencies of
the IDT WinChip C6 processor. The D-TLB shares the 8-entry
unified page directory cache which reduces the TLB miss
penalty. The D-cache is dynamically turned off when not used to
reduce power requirements.

2.3.6 X86 Fetch Unit

One of the most visible size (and frequency)-versus-CPI
tradeoffs made by the IDT WinChip C6 processor is that there is
very limited branch prediction: only a special 8-entry
Call/Return stack. This computer-science heresy (no branch
prediction) helps support the high clock frequency and small die
size but means that, on average, simple branches normally take
slightly more clocks on the IDT WinChip C6 processor than on
other contemporary x86 processors.
However, the clock difference is not as high as might be
expected due to: (1) Return instructions are accurately predicted
with the special Call/Return stack mechanism, (2) the total
performance of complex branches (such as far calls) includes
more than the actual branch time (which is overlapped with the
additional operations on the IDT WinChip C6 processor), and (3)
the short IDT WinChip C6 processor pipeline.
Actually, in a simple pipeline processor like the IDT WinChip
C6 processor, the CPI contribution of branches is not very high.
The main need for branch prediction is in deeply pipelined
superscalar designs where branches happen late in the pipeline
flow and are highly disruptive to the multiple issue strategy. In
the IDT WinChip C6 processor branches happen early in the
short pipeline and invalidate few instructions in the single issue
pipeline.
The most important consideration relative to the IDT WinChip
C6 processor’s lack of branch prediction is that the gain from
branch prediction is relatively small compared to the total time
waiting on bus activity in high-frequency x86 processors. The
IDT WinChip C6 processor design focuses on using its
transistors to minimize this bus wait time: large caches, large
TLBs, smart cache management algorithms, and so forth, as
opposed to a large branch prediction array.

ARCHITECTURE 2-7
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

2.3.7 FP-Unit
In addition to the basic integer execution unit, the IDT WinChip
C6 processor has a separate 80-bit floating-point execution unit
that can execute floating-point instructions in parallel with
integer instructions.
The floating-point unit is designed to maximize clock frequency
and to minimize chip size while providing adequate levels of
floating-point performance for typical desktop use. Some
floating-point instructions are pipelined, but some are only
partially pipelined.
The IDT WinChip C6 processor issues only one instruction per
clock but most integer instructions and most floating-point
instructions can execute in parallel.
The floating-point unit is dynamically turned off when not used
to reduce power requirements.

2.3.8 MMX-Unit
The IDT WinChip C6 processor contains a separate execution
unit for the new MMX-compatible instructions. The MMX
architecture registers are the same as the floating-point
registers, otherwise the MMX execution unit has its own adder,
multiplier and shifter separate from the floating-point unit.
The MMX unit is dynamically turned off when not used to
reduce power requirements.

2.3.9 Bus-Unit
The IDT WinChip C6 processor bus unit provides an external
bus interface compatible with the Pentium processor. In
addition to the expected bus control functions, the bus unit
implements a large page-directory cache to reduce the impact of
TLB misses as well as several special optimizations intended to
reduce cache misses. Four 64-bit write buffers allow internal
execution to proceed overlapped with waiting for external stores
to complete.
The IDT WinChip C6 processor bus unit contains many special
features designed to reduce bus traffic and cache disruption.
Examples include store byte-combining function, cache cast-out
snarfing, “smart” lock management mechanisms, and so forth.

2-8 ARCHITECTURE
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

3 PROGRAMMING INTERFACE

3.1 GENERAL
In general, the IDT WinChip C6 processor is compatible with
both the bus and software-visible architecture of the Intel
Pentium processor with MMX technology. That is, a program
that executes on a Pentium processor should generally execute
on an IDT WinChip C6 processor and produce the same results
(with the exceptions as noted in this datasheet).
The IDT WinChip C6 processor’s Pentium-compatible functions
include:
n All basic X86 instructions, registers, and functions

n All floating-point (numeric processor) instructions, registers

and functions

n All new Pentium processor instructions and registers

(CMPXCHG8B, RDMSR, WRMSR, RDTSC, CPUID, RSM, MOV CR4)

n All basic operating modes: real mode, protect mode, virtual-

8086 mode

n System Management Interrupt (SMI) and the associated

System Management Mode (SMM)

n All interrupt and exception functions

n All debug functions (including the new I/O breakpoint

function)

n All input/output functions

n All tasking functions (TSS, task switch, etc.)

n Processor initialization behavior

The IDT WinChip C6 processor also includes instructions

compatible with the MMX instructions in the Pentium processor
with MMX technology

PROGRAMMING INTERFACE 3-1

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

However, there are some differences between the IDT WinChip

C6 processor and the Pentium processor. These differences fall
into four groups:
n Additional IDT WinChip C6 processor functions.
Examples are memory range registers that allow different
attributes for each range. These additional functions are
provided through Machine Specific Registers such that
compatibility is not affected.
n Implementation-specific differences. Examples are
cache and TLB testing features, and performance monitoring
features that expose the internal implementation features.
These types of functions are incompatible among all different
x86 implementations.—including the Intel486, the Pentium,
and the Pentium Pro processors.
n Omitted functions. Some Pentium processor functions are
not provided on the IDT WinChip C6 processor because they
aren’t used or aren’t needed in the targeted PC systems.
Examples are some specific bus functions such as functional
redundancy checking and performance monitoring.
These types of differences are similar to those among various
versions of the Pentium processor (for example, the mobile
Pentium processor also omits the same bus functions as
omitted by the IDT WinChip C6 processor), and among the
Pentium processor and the AMD-K6 and Cyrix 6x86MX
processors.
n Low-level behavioral differences. A few low-level IDT
WinChip C6 processor functions are different from the
Pentium because the results are (1) documented in the Intel
documentation as undefined, and (2) known to be different
for different x86 implementations (in particular, different
among the Intel i486, the Pentium, and the Pentium Pro
processors). That is, compatibility with the Pentium
processor for these functions is clearly not needed for
software compatibility (or they wouldn’t be different across
different implementations). Where the Pentium and Pentium
Pro processor results differ, the IDT WinChip C6 processor
often provides the Pentium Pro result.
This chapter summarizes the first three types of differences:
additional functions, implementation-specific functions, and
omitted functions. Appendix A contains more details on
machine-specific functions. Appendix B contains details on low-
level differences.

3-2 PROGRAMMING INTERFACE

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

In some areas, we also include comparative information about

the Pentium Pro, AMD-K6, and Cyrix 6x86MX processors. This
information is taken from the data sheets of these products and
has not been verified by Centaur Technology. Our Pentium
processor information, however, is based on detailed testing.

3.2 ADDITIONAL FUNCTIONS

The IDT WinChip C6 processor provides some memory range
management functions. These are similar in concept, but
different in specifics, to memory range registers in the Pentium
Pro, AMD-K6, and Cyrix 6x86MX processors (all of which are
different from each other). These functions are provided via
Machine Specific Registers. Appendix A provides specifics on the
IDT WinChip C6 Machine Specific Registers.

3.3 MACHINE-SPECIFIC FUNCTIONS

3.3.1 General
All x86 processor implementations provide a variety of machine-
specific functions. Examples are cache and TLB testing features,
and performance monitoring features that expose the internal
implementation features. These types of functions are different
and incompatible among all different x86 implementations—
including the Intel i486, the Pentium, and the Pentium Pro
processors, and between these processors and competitive
processors from Cyrix and AMD. The Intel documentation
clearly identifies these types of functions as machine-specific
and warns of possible changes in new implementations.
This section describes the IDT WinChip C6 processor machine-
specific functions that are most likely used by software and
compares them to related processors where applicable. Appendix
A describes the IDT WinChip C6 processor machine-specific
registers (MSRs).
This section covers those features of Pentium-compatible
processors that are used to commonly identify and control
processor features. All Pentium-compatible processors have the
same mechanisms, but the bit-specific data values often differ.

PROGRAMMING INTERFACE 3-3

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

3.3.2 CPUID Instruction

The CPUID instruction is available on all contemporary x86
processors. The CPUID instruction has two basic functions
requested via the EAX register. The first function returns a
vendor identification string in registers EBX, ECX and EDX.
The second CPUID function returns an assortment of bits in
EAX and EDX that identify the chip version and describe the
specific features available.
The EAX:EBX:ECX:EDX return values of the CPUID
instruction executed with EAX == 0 are:

Table 3-1
Register[bits] - C6 P54 P55 K6 M2
Meaning
EAX 1 1 1
(highest EAX input
value understood
by CPUID)
EBX:EDX:ECX “Centaur “Genuine “Genuine “Authentic “Cyrix
(vendor ID string) Hauls” Intel” Intel” AMD” Instead”

The EAX return values of the CPUID instruction executed with

EAX == 1 are:

Table 3-2
EAX Bits - C6 P54 P55 K6 M2
Meaning
3:0 - Stepping ID
7:4 - Model ID Same as the return value in EDX after Reset
(see next section)
11:8 - Family ID

13:12 - Type ID

3-4 PROGRAMMING INTERFACE

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

The EDX return values of the CPUID instruction with EAX == 1

are:

Table 3-3
EAX Bits - Meaning C6 P54 P55 K6 M2 Notes
0 - FP present 1 1 1 1 1
1 - VM86 Extensions (VME) 0 1 1 1 0 1
2 - Debugging Extensions 1 1 1 1 1
3 - Page Size Extensions (4MB) 0 1 1 1 0 1
4 - Time Stamp Counter (TSC) 1 1 1 1 1 2
supported
5 - Model Specific Registers 1 1 1 1 1 3
present
6 - PAE supported (P6 Function) 0 0 0 0 0 4
7 - Machine Check Exception 1/0 1 1 1 0 5
8 - CMPXCHG8B instruction 0/1 1 1 1 1 6
9 - APIC supported 0 1 1 0 0 7
10:11 -Reserved
12- Memory Range Registers 0 0 0 0 0 8
13 - PTE Global Bit supported 0 0 0 0 1 4
14- Machine Check Architecture 0 0 0 0 0 4
supported
15- Conditional Move supported 0 0 0 0 1 4
16:22 - Reserved
23 - MMX supported 1/0 0 1 1 1 98
24:31 - Reserved

PROGRAMMING INTERFACE 3-5

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

Notes On CPUID Feature Flags

General: an “x/y” entry means that the default setting of this bit
is x but the bit (and the underlying function) can be set to y
using the FCR MSR.
1. These “Appendix H” functions are not provided on the IDT
WinChip C6 processor since they are not used by the
target operating systems. They are also not provided on
the Cyrix 6x86MX processor.
2. The IDT WinChip C6 processor implementation varies
slightly from that of the Pentium processor in a way that
should have no practical impact.
3. Every system has different MSRs—addresses and
contents.
4. This is a function introduced with the Pentium Pro
processor and is generally not provided on Pentium-
compatible processors.
5. The Machine Check exception is defined by Intel
documentation as machine-specific. The IDT WinChip C6
processor’s Machine Check has slightly different specifics
than the Pentium processor’s Machine Check function.
The Machine Check support can be enabled or disabled by
a bit in the FCR MSR. The CPUID bit reports the current
setting of this enable control.
6. The Pentium processor-compatible CMPXCHG8B
instruction is provided and always enabled. However, the
default for the corresponding CPUID function bit is 0 (due
to a bug found in Windows NT). This default can be
changed via a bit in the FCR MSR.
7. This is a Intel-specific multiprocessing function. None of
the other Pentium-compatible processors provide this
function since it has no utility in the target system
environment.
8. The IDT WinChip C6 and Cyrix 6x86MX processors have
memory range registers, but the specifics are not
compatible with the P6 MRRs. The Pentium processor has
no memory range registers.

3-6 PROGRAMMING INTERFACE

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

9. The IDT WinChip C6 processor’s MMX-technology

compatible instruction support can be enabled or disabled
by a bit in the FCR. The CPUID bit reports the current
setting of this enable control.

3.3.3 Extended CPUID Instruction Functions

The IDT WinChip C6 processor provides several machine-
specific features. Some of these features are compatible with
those provided by P55 and are identified by the standard
CPUID function EAX=1.
Other machine-specific features described in this datasheet
have no P55 equivalent. These features are controlled by IDT
WinChip C6 MSRs.
System software must not assume that all future processors in
the IDT WinChip C6 family will implement all of the same
machine-specific features or even that these features will be
implemented in a backward-compatible manner. In order to
determine if the processor supports particular machine-specific
features, system software should follow this procedure:
1. Identify the processor as an IDT WinChip C6 by using
CPUID with EAX=0
2. Execute the CPUID instruction with extended function
EAX = C000_0000h. This function is unique to IDT
WinChip family processors.
• If the output in EAX is C000_0000h (unchanged) then
the IDT WinChip C6 supports the machine-specific
features as described in this datasheet.
• If EAX output is something other than C000_0000h
then the processor is a later version of the WinChip C6
and system software must not attempt to activate any
machine-specific feature.

PROGRAMMING INTERFACE 3-7

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

3.3.4 EDX Value After Reset.

As for other x86 processors, after reset the EDX register holds a
component identification number as follows:
31:14 13:12 11:8 7:4 3:0

EDX reserved Type ID Family ID Model ID Stepping ID

18 2 4 4 4

The specific values for the various IDT WinChip C6 processor

types are:

Processor Type ID Family ID Model ID Stepping ID

C6 0 5 4 varies

For comparison, following are the values for other X86

processors:
Processor Type ID Family ID Model ID Stepping ID
P54C 0 5 2 varies
P55 0 5 4 varies
Cyrix 6x86MX 0 6 0 varies
AMD-K6 0 6 6-9 varies

3-8 PROGRAMMING INTERFACE

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

3.3.5 CR4
Control register 4 (CR4) is a new feature of the Pentium
processor that controls some of its advanced features. The IDT
WinChip C6 processor provides a CR4 with the following
specifics:

CR4 Bits - Meaning C6 P54 P55 K6 M2 Notes

0: VME: Enables VME 0 0/1 0/1 0/1 0 1
feature
1: PVI: Enables PVI 0 0/1 0/1 0/1 0 1
feature
2: TSD: Makes RDTSC 0/1 0/1 0/1 0/1 0/1
inst privileged
3: DE: Enables I/O 0/1 0/1 0/1 0/1 0/1
breakpoints
4: PSE: Enables 4-MB 0 0/1 0/1 0/1 0 1
pages
5: PAE: Enables addr r r r r r 2
extensions
6: MCE: Enables 0/1 0/1 0/1 0/1 0 3
machine check
exception
7: PGE: Enables global r r r r 0/1 2
page feature
8: PCE: Enables 0/1 r 0/1 r 0/1
RDPMC for all
levels
31:9 - reserved r r r r r

PROGRAMMING INTERFACE 3-9

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

Notes On CR4
General: a “0/1” means that the default setting of this bit is 0
but the bit can be set to (1). A “0” means that the bit is always 0;
it cannot be set. An “r” means that this bit is reserved. It
appears as a 0 when read, and a GP exception is signaled if an
attempt is made to write a 1 to this bit.
1. The IDT WinChip C6 processor does not provide this
“Appendix H” function and this CR4 bit cannot be set.
However, no GP exception occurs if an attempt is made to
set this bit. The Cyrix 6x86MX processor also does not
provide this function.
2. This is a Pentium-Pro processor function that is typically
not provided on P55-compatible processor.
3. The IDT WinChip C6 processor Machine Check has
slightly different specifics than the P54C Machine Check
function

3.3.6 Machine-Specific Registers

The IDT WinChip C6 processor implements the Pentium family
concept of Machine Specific Registers (MSRs). RDMSR and
WRMSR instructions are provided and the CPUID instruction
identifies that the IDT WinChip C6 processor supports MSRs.
However, the IDT WinChip C6 processor MSRs are different
from the Pentium and Pentium Pro processors (which are
different from each other, and from the Cyrix 6x86MX and
AMD-K6 processors).
In general, the MSRs have no usefulness to application or
operating system software and are not used. (This is to be
expected since the MSRs are different on each processor).
Appendix A contains a detailed description of the IDT WinChip
C6 processor’s MSRs.

3.4 OMITTED FUNCTIONS

This section summarizes those functions that are included in
some Pentium processor versions, but are not in the IDT
WinChip C6 processor.

3-10 PROGRAMMING INTERFACE

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

3.4.1 Pentium Appendix H Enhancements

The infamous Appendix H functions are those Pentium
functions that are documented in Appendix H (Advanced
Functions) of Volume 3 of the Pentium Processor Family
Developer’s Manual.
Unfortunately, Appendix H is only available to those with the
“appropriate non-disclosure agreements in place”. However,
most of these functions are now publicly documented in the
Pentium Pro processor documentation.
The Appendix H features are identified as “optional in future
are specifically identified as being supported or
not by the CPUID instruction.
Due to the limited utility of these advanced functions (they are
complex operating system functions), there are few programs
that utilize these features. In particular, these functions are
either not used at all, or are conditionally used if present, by
Microsoft desktop operating systems.
The IDT WinChip C6 processor does not provide the following
Appendix H functions.

Virtual Memory Enhancements (4-MB Pages).

These Pentium processor enhancements provide the ability to
optionally define 4-MB virtual memory pages in addition to the
usual 4-KB page size. A bit in the feature identification return
from the CPUID instruction indicates whether this feature is
present or not. This enhancement is not provided on the IDT
WinChip C6 processor since it is not used by the target
operating systems: Windows 95 and Windows 98. Note that this
function is also not provided on the Cyrix 6x86MX processor.

Virtual-8086 Mode Enhancements (VME)

These Pentium processor enhancements provide potential
performance improvements to mode-switching operations while
operating in VM86 mode. A bit in the feature identification
return from the CPUID instruction indicates whether this
feature is present or not. This enhancement is not provided on
the IDT WinChip C6 processor since it is not used by the target
Microsoft operating systems. Note that this function is also not
provided on the Cyrix 6x86MX processor.

PROGRAMMING INTERFACE 3-11

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

3.4.2 Other Functions

The IDT WinChip C6 processor also omits the software interface
to the Intel-proprietary symmetric multiprocessing support:
APIC. This bus function is omitted since the target market for
the IDT WinChip C6 processor is portables and typical desktop
systems (which do not support APIC multiprocessing).
A bit in the feature identification return from the CPUID
instruction indicates whether this feature is present or not. This
enhancement is not provided on the IDT WinChip C6 processor
(as it is not on the mobile Pentium processor and on the AMD-
K6 and Cyrix 6x86MX processors).

3-12 PROGRAMMING INTERFACE

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

4 HARDWARE INTERFACE

4.1 BUS INTERFACE

The IDT WinChip C6 processor bus interface is compatible with
the Pentium processor and the Pentium processor with MMX
technology. This behavior is specified in Pentium Processor
Family Developer’s Manual.
The majority of the pins within the bus interface are involved
with the physical memory and I/O interface. These pins and
this interface perform the same functions as in the Pentium
processor. The remaining pins are power and ground pins, test
and debug support pins and various ancillary control functions.
Most of these pins are identical to the Pentium processor.
Others are associated with functions that behave slightly
differently from the Pentium processor on the IDT WinChip C6
processor. Still others behave differently among the various
versions of the Pentium processor, and thus require clarification
on the IDT WinChip C6 processor. Lastly there are several
Pentium processor functions which are completely omitted on
the IDT WinChip C6.

4.1.1 Differences
The areas where the IDT WinChip C6 processor differs from the
Pentium processor are not anticipated to cause operational
compatibility issues. These differences are:
n Bus Frequency Control
n Machine Check Exceptions on BUSCHK# and PEN#
n Drive Strengths
n Probe Mode / JTAG / TAP Port (see Test and Debug Section)

HARDWARE INTERFACE 4-1

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

Bus Frequency Control

Because the IDT WinChip C6 is designed to support several
frequencies higher than 233 MHz, the IDT WinChip C6
processors interpret the bus frequency control pin slightly
differently from the Pentium processor’s specification. The IDT
WinChip C6’s interpretation is shown in Table 4-1. Note that
the changes redefine fractional bus frequency multiples (which
the IDT WinChip C6 does not support) to higher multiples. A
new pin, BF2 (W-35), is also supported to allow for higher
frequencies. The IDT WinChip C6 pulls up all three of the BF
pins so that the default is 4x multiplier.

Table 4-1. Bus Frequency Ratios

BF2 BF1 BF0 IDT WinChip C6 P55 Clock

Clock Ratio Ratio
1 0 0 reserved 5/2x
1 0 1 3x 3x
1 1 0 2x 2x
1 1 1 4x 3/2x
0 0 0 reserved
0 0 1 5x
0 1 0 4x
0 1 1 reserved

Machine Check Exceptions on BUSCHK# & PEN#

As in the Pentium processor, the BUSCHK# interrupt causes a
Machine Check exception or is ignored based on CR4.MCE. The
difference is that the semantics of Machine Check exception are
slightly different on a IDT WinChip C6 processor from a
Pentium processor (a IDT WinChip C6 processor doesn’t save
and report the bus address and cycle data). See section 4.4 for
further description of Machine Check.

Drive Strength
Desktop Pentium processors have three driver strengths that
can be selected at Reset for certain pins (for example ADS#).
The driver strength is selected by the BRDYC# and BUSCHK#
pins when sampled at RESET deassertion

4-2 HARDWARE INTERFACE

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

The IDT WinChip C6 processor has only two driver strengths:

Table 4-2. Drive Strengths

BUSCHK# BRDYC# IDT WinChip C6 Driver P54C Driver
0 0 Medium Strong
0 1 Medium Medium
1 0 Typical Typical
1 1 Typical Typical
Only ADS#, A[20:3], HITM# and W/R# are configurable. All
other drivers are typical strength. The AC characteristics of
both drive strengths are described in Chapter 5, Electrical
Specifications.

4.1.2 Clarifications
Power Supply Voltage
The IDT WinChip C6 processor operates with a unified power
plane. Depending on the version, the processor requires either
3.3 Volts or 3.52 Volts at its VCC inputs.
The IDT WinChip C6 package is compatible with Socket 7, in
that the VCC2DET# pin is internally no-connected. Flexible
socket 7 motherboards can use the fact that VCC2DET# is not
internally connected to force the motherboard core and pad
regulators to produce the same voltage.

5V Tolerance
Like the P55, the IDT WinChip C6 processor’s CLK input is not
5 Volt tolerant. It should be driven by a 3.3 Volt device.

4.1.3 Omissions
Advanced Peripheral Interrupt Controller (APIC)
The APIC is not supported by the IDT WinChip C6. The APIC
pins (PICCLK, PICD0, and PICD1) are classified as reserved,
and should not be connected on the motherboard.
(The APIC is also not supported in the mobile Pentium
processor, the Cyrix 6x86MX and AMD-K6 processors.)

HARDWARE INTERFACE 4-3

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

Dual Processor Interface

The IDT WinChip C6 processors do not support the dual
processor interface. The associated pins (D/P#, PBGNT#,
PBREQ#, PHIT#, and PHITM#) are classified as reserved, and
should not be connected on the motherboard.
(The DP interface is also not supported in the mobile Pentium
processor, the Cyrix 6x86MX and AMD K6.)

Functional Redundancy Checking Mode

The IDT WinChip C6 processors do not support the functional
redundancy checking mode. The FRCMC# pin is classified as
reserved, and should not be connected on the motherboard
(The functional redundancy checking mode is also not supported
in the mobile Pentium processors, the Pentium processors with
MMX technology, and the Cyrix 6x86MX.)

Breakpoint and Performance Monitoring Signals

The IDT WinChip C6 processors internally support instruction
and data breakpoints. However, the IDT WinChip C6 does not
support the Pentium processor’s external indication of
breakpoint matches via the BP3-BP0 pins. Similarly, the IDT
WinChip C6 contains performance monitoring hooks internally,
but it does not support the Pentium processor’s external
indication of performance monitoring events on PM1-PM0. The
associated pins are unconnected on the IDT WinChip C6
package.

4.2 SIGNAL SUMMARY

Table 4-3 summarizes the bus interface signals of a Pentium
and which signals are provided on a IDT WinChip C6 processor:
an ‘×’ in each processor’s column indicates that the pin is
supported by that processor.

4-4 HARDWARE INTERFACE

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

Table 4-3. Signal Summary

Signal Type P55 P55 P54C P54C C6
(TCP) (PPGA) (TCP) (PPGA) (CPGA)
A20M# I × × × × ×
A31-A3 I/O × × × × ×
ADS# O × × × × ×
ADSC# O × × ×
AHOLD I × × × × ×
AP I/O × × × × ×
APCHK# I × × × × ×
APICEN/PICD I/O × ×
1
BE7#-BE0# O × × × × ×
APICID[3:0] - × × ×
BF[2:0] I × × × × ×
BOFF# I × × × × ×
BP[3:2] O × × × ×
PM/BP[1:0]
BRDY# I × × × × ×
BRDYC# I × × ×
BREQ O × × × × ×
BUSCHK# I × × × × ×
CACHE# O × × × × ×
CLK I × × × × ×
CPUTYP I × ×
D/C# O × × × × ×
D63-D0 I/O × × × × ×
D/P# - × ×
DP7-DP0 I/O × × × × ×
DPEN# - - × ×
PICD0
EADS# I × × × × ×
EWBE# I × × × × ×

HARDWARE INTERFACE 4-5

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

Signal Type P55 P55 P54C P54C C6

(TCP) (PPGA) (TCP) (PPGA) (CPGA)
FERR# O × × × × ×
FLUSH# I × × × × ×
FRCMC# - × ×
HIT# I × × × × ×
HITM# I × × × × ×
HLDA O × × × × ×
HOLD I × × × × ×
IERR# O × × × × ×
IGNNE# I × × × × ×
INIT I × × × × ×
INV I × × × × ×
KEN# O × × × × ×
INTR I × × × × ×
NMI I × × × × ×
LOCK# O × × × × ×
M/IO# O × × × × ×
NA# I × × × × ×
PBGNT# - × ×
PBREQ# - × ×
PCD O × × × × ×
PCHK# O × × × × ×
PEN# I × × × × ×
PHIT# - × ×
PHITM# - × ×
PICCLK - × ×
PRDY O × × × × ×
PWT O × × × × ×
R/S# I × × × × ×
RESET I × × × × ×
SCYC O × × × × ×

4-6 HARDWARE INTERFACE

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

Signal Type P55 P55 P54C P54C C6

(TCP) (PPGA) (TCP) (PPGA) (CPGA)
SMI# I × × × × ×
SMIACT# O × × × × ×
STPCLK# I × × × × ×
TCK I × × × × ×
TDI I × × × × ×
TDO O × × × × ×
TMS I × × × × ×
TRST# I × × × × ×
VCC2DET# - × × ×
W/R# O × × × × ×
WB/WT# O × × × × ×

4.3 POWER MANAGEMENT

The IDT WinChip C6 processor provides both static and
dynamic power management.

4.3.1 Static Power Management

The IDT WinChip C6 processor supports the five power
management modes of the Pentium processor: NORMAL state,
STOP CLOCK state, STOP GRANT state, STOP CLOCK
SNOOP state, and AUTOHALT state. These are described in
the Pentium Family Developer’s Manual.

4.3.2 Dynamic Power Management

The IDT WinChip C6 processor uses dynamic power
management techniques to reduce power consumption in the
NORMAL state. In NORMAL state, the on-chip arrays,
selected datapaths, and the associated control logic are powered
down when not in use.

HARDWARE INTERFACE 4-7

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

4.4 TEST & DEBUG

4.4.1 Machine Check

IDT WinChip C6 Processors provide an Machine Check
exception function (INT 18) that is slightly different than the
Pentium processor or Pentium Pro processor Machine Check
function (which are different from each other, of course). These
differences are reasonable and expected since Intel
documentation specifies that the Machine Check architecture is
processor-specific.
In both the Pentium processor and IDT WinChip C6 processor,
the Machine Check exception must be enabled by setting the
MCE bit in CR4. If not enabled, the conditions (below) causing a
Machine Check are ignored and no processor action is taken.
Both the IDT WinChip C6 processor and Pentium processor
cause a Machine Check, if enabled, when:
n BUSCHK# is asserted
n PEN# is asserted and a data parity error is detected (PCHK#
is asserted)
The differences between the IDT WinChip C6 processor and the
Pentium processor are:
n The Pentium processor reports specifics about the bus cycle
in MSR’s 0 and 1. The IDT WinChip C6 processor does not
provide this bus-cycle data.
n The IDT WinChip C6 processor default behavior for
internally detected processor errors is (like the Pentium
processor) to assert IERR and (normally) perform to a
Shutdown bus cycle. However, if the EMCIE bit in the FCR
is set, then internal errors on a IDT WinChip C6 processor
cause a Machine Check exception.

4.4.2 BIST
A Built-in Self Test (BIST) can be requested as part of the IDT
WinChip C6 processor reset sequence using exactly the same
mechanism as used on the Pentium processor (INIT asserted as
RESET deasserted).

4-8 HARDWARE INTERFACE

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

The IDT WinChip C6 processor BIST performs the following

general functions:
n A hardware-implemented exhaustive test of (1) all internal
microcode ROM, and (2) the X86 instruction decode,
instruction generation and entry point generation logic.
n An extensive microcode test of all internal registers and
datapaths.
n An extensive microcode test of data and instruction caches,
their tags, and associated TLB’s.
BIST requires about two million internal clocks.

EAX Value After Reset

The results of a BIST is indicated by a code in EAX. Normally
EAX is zero after reset. If a BIST is requested as part of the
Reset sequence, EAX contains the BIST results. A 0 in EAX
after BIST Reset means that no failures were detected. Any
value other than zero indicates an error has occurred during
BIST.

4.4.3 Internal Error Detection

During normal execution, the IDT WinChip C6 processor detects
parity errors in both caches. In addition, certain “impossible”
internal states are detected by microcode. These errors are
normally reported via the same mechanism as in the Pentium
processor: the IERR bus signal is asserted and (normally) a
Shutdown occurs. (The Shutdown can be suppressed on both
processors via a control bit in the TR1 MSR.)
Alternatively, an optional feature (a control bit in the FCR
MSR) allows internal errors to be reported as Machine Check
exceptions.

4.4.4 JTAG
The IDT WinChip C6 processor has a JTAG scan interface
which is used for test functions and the proprietary Debug Port.
However, unlike the Pentium processor, the IDT WinChip C6
processor does not provide a fully compatible IEEE 1149.1 JTAG
function. In particular, the boundary scan function is not
provided.
From a practical user viewpoint, JTAG does not exist and the
associated pins (TCK, and so forth) should not be used.

HARDWARE INTERFACE 4-9

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

4.4.5 Debug Port

The Pentium processor (and other processors such as the AMD-
K6) have a proprietary Debug Port which uses the JTAG scan
mechanism to control internal debug features (“probe mode”).
These interfaces are not documented and are available (if at all)
only under a non-disclosure agreement.
Similarly, the IDT WinChip C6 processor has an undocumented
and proprietary debug interface.

4-10 HARDWARE INTERFACE

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

5 ELECTRICAL SPECIFICATIONS

5.1 AC TIMING TABLES FOR 75-MHZ BUS

Table 5-1. AC Specifications for 75-Mhz Bus
SYMBOL PARAMETER MIN MAX UNIT FIGURE NOTES

f CLK Frequency 37.5 75 MHz

t1a CLK Period 13 1/3 26 2/3 ns

t1b CLK Period Stability ±250 ps Adjacent

Clocks

t2 CLK High Time 4.0 ns 2V

t3 CLK Low Time 4.0 ns 0.8V

t4 CLK Fall Time 0.15 1.5 ns 2V-0.8V

t5 CLK Rise Time 0.15 1.5 ns 0.8V-2V

t6a PWT, PCD, CACHE# Valid Delay 1.0 7.0 ns (1)

t6b AP Valid Delay 1.0 8.5 ns (1)

t6c BE0-7#, LOCK# Valid Delay 0.9 7.0 ns (1)

t6d ADS# Valid Delay 0.8 6.0 ns (1)

t6e ADC#, D/C#, W/R#, SCYC Valid Delay 0.8 7.0 ns (1)

t6f M/IO# Valid Delay 0.8 5.9 ns (1)

t6g A3-A16 Valid Delay 0.5 6.3 ns (1)

t6h A17-A31 0.6 6.3 ns (1)

t7 ADS#, ADSC#, AP, A3-A31, PWT, 10.0 ns (1)

PCD, BE0-7#, M/IO#, D/C#, W/R#,
CACHE#, SCYC, LOCK# Float Delay

t8a APCHK#, IERR#, FERR# Valid Delay 1.0 8.3 ns (1)

t8b PCHK# Valid Delay 1.0 7.0 ns (1)

t9a BREQ Valid Delay 1.0 8.0 ns (1)

t9b SMIACT# Valid Delay 1.0 7.3 ns (1)

t9c HLDA Valid Delay 1.0 6.8 ns (1)

t10a HIT# Valid Delay 1.0 6.8 ns (1)

t10b HITM# Valid Delay 0.7 6.0 ns (1)

t11b PRDY Valid Delay 1.0 8.0 ns (1)

t12 D0-D63, DP0-7 Valid Delay 1.3 7.5 ns (1)

t13 D0-D63, DP0-7 Float Delay 10.0 ns (2)

t14 A5-A31 Setup Time 3.3 ns

t15 A5-A31 Hold Time 1.0 ns

ELECTRICAL SPECIFICATIONS 5-1

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

SYMBOL PARAMETER MIN MAX UNIT FIGURE NOTES

t16a INV, AP Setup Time 3.3 ns

t16b EADS# Setup Time 3.3 ns

t17 EADS#, INV, AP Hold Time 1.0 ns

t18a KEN# Setup Time 3.0 ns

t18b NA#, WB/WT#, NA# Setup Time 3.3 ns

t19 KEN#, WB/WT#, NA# Hold Time 1.0 ns

t20 BRDY#, BRDYC# Setup Time 3.3 ns

t21 BRDY#, BRDYC# Hold Time 1.0 ns

t22 AHOLD, BOFF# Setup Time 3.3 ns

t23 AHOLD, BOFF# Hold Time 1.0 ns

t24a BUSCHK#, EWBE#, HOLD Setup Time 3.3 ns

t24b PEN# Setup Time 3.3 ns

t25a BUSCHK#, EWBE#, PEN# Hold Time 1.0 ns

t25b HOLD Hold Time 1.5 ns

t26 A20M#, INTR, STPCLK# Setup Time 3.3 ns (4)

t27 A20M#, INTR, STPCLK# Hold Time 1.0 ns (5)

t28 INIT, FLUSH#, NMI, SMI#, IGNNE# 3.3 ns (4)

Setup Time

t29 INIT, FLUSH#, NMI, SMI#, IGNNE# 1.0 ns (5)

Hold Time

t30 INIT, FLUSH#, NMI, SMI#, IGNNE# 2.0 CLK’s (6)

Pulse Width (Async)

t31 R/S# Setup Time 3.3 ns

t32 R/S# Hold Time 1.0 ns

t33 R/S# Pulse Width (Async) 2.0 ns

t34 D0-63, DP0-7 Read Data Setup Time 2.8 ns

t35 D0-63, DP0-7 Read Data Hold Time 1.5 ns

t36 RESET Setup Time 3.3 ns (4)

t37 RESET Hold Time 1.0 ns (5)

t38 RESET Pulse Width, Vcc & CLK Stable 15.0 CLK’s

t39 RESET Pulse Width after Vcc & CLK 1.0 ms Power
Stable Up

t40 RESET Configuration Signals (INIT, 3.3 ns

FLUSH#) Setup Time

t41 RESET Configuration Signals (INIT, 1.0 ns

FLUSH#) Hold Time

t42a RESET Configuration Signals (INIT, 2.0 CLK’s To

FLUSH#) Setup Time (Async) RESET
Falling
Edge

5-2 ELECTRICAL SPECIFICATIONS

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

SYMBOL PARAMETER MIN MAX UNIT FIGURE NOTES

t42b RESET Configuration Signals (INIT, 2.0 CLK’s To

FLUSH#, BRDYC#, BUSCHK#) Hold RESET
Time (Async) Falling
Edge

t42c RESET Configuration Signals 3.0 CLK’s To

(BRDYC#, BUSCHK#) Setup Time RESET
(Async) Falling
Edge

t42d RESET Configuration Signals 1.0 ns To

(BRDYC#, BUSCHK#) Hold Time RESET
(Sync) Falling
Edge

t43a BF0, BF1, BF2 Setup Time 1.0 ms To

RESET
Falling
Edge

t43b BF0, BF1, BF2 Hold Time 2.0 CLK’s To

RESET
Falling
Edge

Notes:
1. CL = 0 pF
2. Not 100% tested. Guaranteed by design and characterization
3. All outputs are glitch free signals, guaranteed to rise and fall monotonically when driven into
capacitive loads. Most system loads must be treated as transmission lines. Depending on the
length of the transmission line, loading and impedance mismatches, the signal may not rise or
fall monotonically at a given point along the transmission line.
4. Setup time must be met to guarantee sampling on a given processor clock. Signals may be
driven asynchronously, but, if so, are not guaranteed to be sampled by a specific clock edge.
5. Hold time must be met to guarantee sampling on a given processor clock. Signals may be driven
asynchronously, but, if so, are not guaranteed to be sampled by a specific clock edge.
6. To guarantee proper recognition, signal must be deasserted for two or more processor clocks
when driven asynchronously. When driven synchronously, signal must be deasserted for one
processor clock to be recognized.

5.2 AC TIMING TABLES FOR 66-MHZ BUS

Table 5-2. AC Specifications for 66-MHz Bus

SYMBOL PARAMETER MIN MAX UNIT FIGURE NOTES

f CLK Frequency 33 1/3 66 2/3 MHz

t1a CLK Period 15.0 30.0 ns

t1b CLK Period Stability ±250 ps Adjacent

Clocks

t2 CLK High Time 4.0 ns

t3 CLK Low Time 4.0 ns

ELECTRICAL SPECIFICATIONS 5-3

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

SYMBOL PARAMETER MIN MAX UNIT FIGURE NOTES

t4 CLK Fall Time 0.15 1.5 ns

t5 CLK Rise Time 0.15 1.5 ns

t6a PWT, PCD, CACHE# Valid Delay 1.0 7.0 ns (1)

t6b AP Valid Delay 1.0 8.5 ns (1)

t6c BE0-7#, LOCK# Valid Delay 0.9 7.0 ns (1)

t6d ADS# Valid Delay 0.8 6.0 ns (1)

t6e ADC#, D/C#, W/R#, SCYC Valid Delay 0.8 7.0 ns (1)

t6f M/IO# Valid Delay 0.8 5.9 ns (1)

t6g A3-A16 Valid Delay 0.5 6.3 ns (1)

t6h A17-A31 0.6 6.3 ns (1)

t7 ADS#, ADSC#, AP, A3-A31, PWT, 10.0 ns (1)

PCD, BE0-7#, M/IO#, D/C#, W/R#,
CACHE#, SCYC, LOCK# Float Delay

t8a APCHK#, IERR#, FERR# Valid Delay 1.0 8.3 ns (1)

t8b PCHK# Valid Delay 1.0 7.0 ns (1)

t9a BREQ Valid Delay 1.0 8.0 ns (1)

t9b SMIACT# Valid Delay 1.0 7.3 ns (1)

t9c HLDA Valid Delay 1.0 6.8 ns (1)

t10a HIT# Valid Delay 1.0 6.8 ns (1)

t10b HITM# Valid Delay 0.7 6.0 ns (1)

t11b PRDY Valid Delay 1.0 8.0 ns (1)

t12 D0-D63, DP0-7 Valid Delay 1.3 7.5 ns (1)

t13 D0-D63, DP0-7 Float Delay 10.0 ns (2)

t14 A5-A31 Setup Time 6.0 ns

t15 A5-A31 Hold Time 1.0 ns

t16a INV, AP Setup Time 5.0 ns

t16b EADS# Setup Time 5.0 ns

t17 EADS#, INV, AP Hold Time 1.0 ns

t18a KEN# Setup Time 5.0 ns

t18b NA#, WB/WT#, NA# Setup Time 4.5 ns

t19 KEN#, WB/WT#, NA# Hold Time 1.0 ns

t20 BRDY#, BRDYC# Setup Time 5.0 ns

t21 BRDY#, BRDYC# Hold Time 1.0 ns

t22 AHOLD, BOFF# Setup Time 5.5 ns

t23 AHOLD, BOFF# Hold Time 1.0 ns

t24a BUSCHK#, EWBE#, HOLD Setup Time 5.0 ns

t24b PEN# Setup Time 4.8 ns

5-4 ELECTRICAL SPECIFICATIONS

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

SYMBOL PARAMETER MIN MAX UNIT FIGURE NOTES

t25a BUSCHK#, EWBE#, PEN# Hold Time 1.0 ns

t25b HOLD Hold Time 1.5 ns

t26 A20M#, INTR, STPCLK# Setup Time 5.0 ns (4)

t27 A20M#, INTR, STPCLK# Hold Time 1.0 ns (5)

t28 INIT, FLUSH#, NMI, SMI#, IGNNE# 5.0 ns (4)

Setup Time

t29 INIT, FLUSH#, NMI, SMI#, IGNNE# 1.0 ns (5)

Hold Time

t30 INIT, FLUSH#, NMI, SMI#, IGNNE# 2.0 CLK’s

Pulse Width (Async)

t31 R/S# Setup Time 5.0 ns

t32 R/S# Hold Time 1.0 ns

t33 R/S# Pulse Width (Async) 2.0 ns

t34 D0-63, DP0-7 Read Data Setup Time 2.8 ns

t35 D0-63, DP0-7 Read Data Hold Time 1.5 ns

t36 RESET Setup Time 5.0 ns

t37 RESET Hold Time 1.0 ns

t38 RESET Pulse Width, Vcc & CLK Stable 15.0 CLK’s

t39 RESET Pulse Width after Vcc & CLK 1.0 ms Power
Stable Up

t40 RESET Configuration Signals (INIT, 5.0 ns

FLUSH#) Setup Time

t41 RESET Configuration Signals (INIT, 1.0 ns

FLUSH#) Hold Time

t42a RESET Configuration Signals (INIT, 2.0 CLK’s To

FLUSH#) Setup Time (Async) RESET
Falling
Edge

t42b RESET Configuration Signals (INIT, 2.0 CLK’s To

FLUSH#, BRDYC#, BUSCHK#) Hold RESET
Time (Async) Falling
Edge

t42c RESET Configuration Signals 3.0 CLK’s To

(BRDYC#, BUSCHK#) Setup Time RESET
(Async) Falling
Edge

t42d RESET Configuration Signals 1.0 ns To

(BRDYC#, BUSCHK#) Hold Time RESET
(Sync) Falling
Edge

t43a BF0, BF1, BF2 Setup Time 1.0 ms To

RESET
Falling
Edge

t43b BF0, BF1, BF2 Hold Time 2.0 CLK’s To

RESET

ELECTRICAL SPECIFICATIONS 5-5

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

SYMBOL PARAMETER MIN MAX UNIT FIGURE NOTES

Falling
Edge

See notes after Table 5-1.

5.3 AC TIMING TABLES FOR 60-MHZ BUS

Table 5-3. AC Specifications for 60-MHz Bus

SYMBOL PARAMETER MIN MAX UNIT FIGURE NOTES

f CLK Frequency 30 60 MHz

t1a CLK Period 16 2/3 33 1/3 ns

t1b CLK Period Stability ±250 ps Adjacent

Clocks

t2 CLK High Time 4.0 ns

t3 CLK Low Time 4.0 ns

t4 CLK Fall Time 0.15 1.5 ns

t5 CLK Rise Time 0.15 1.5 ns

t6a PWT, PCD, CACHE# Valid Delay 1.0 7.0 ns

t6b AP Valid Delay 1.0 8.5 ns

t6c BE0-7#, LOCK# Valid Delay 0.9 7.0 ns

t6d ADS# , ADC#, D/C#, M/IO#, W/R#, 0.8 7.0 ns

SCYC Valid Delay

t6e A3-A16 Valid Delay 0.5 6.3 ns

t6f A17-A31 0.6 6.3 ns

t7 ADS#, ADSC#, AP, A3-A31, PWT, 10.0 ns

PCD, BE0-7#, M/IO#, D/C#, W/R#,
CACHE#, SCYC, LOCK# Float Delay

t8a APCHK#, IERR#, FERR# Valid Delay 1.0 8.3 ns

t8b PCHK# Valid Delay 1.0 7.0 ns

t9a BREQ, HLDA Valid Delay 1.0 8.0 ns

t9b SMIACT# Valid Delay 1.0 7.6 ns

t10a HIT# Valid Delay 1.0 8.0 ns

t10b HITM# Valid Delay 0.7 6.0 ns

t11b PRDY Valid Delay 1.0 8.0 ns

t12 D0-D63, DP0-7 Valid Delay 1.3 7.5 ns

t13 D0-D63, DP0-7 Float Delay 10.0 ns

t14 A5-A31 Setup Time 6.0 ns

t15 A5-A31 Hold Time 1.0 ns

5-6 ELECTRICAL SPECIFICATIONS

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

SYMBOL PARAMETER MIN MAX UNIT FIGURE NOTES

t16a INV, AP Setup Time 5.0 ns

t16b EADS# Setup Time 5.5 ns

t17 EADS#, INV, AP Hold Time 1.0 ns

t18a KEN# Setup Time 5.0 ns

t18b NA#, WB/WT#, NA# Setup Time 4.5 ns

t19 KEN#, WB/WT#, NA# Hold Time 1.0 ns

t20 BRDY#, BRDYC# Setup Time 5.0 ns

t21 BRDY#, BRDYC# Hold Time 1.0 ns

t22 AHOLD, BOFF# Setup Time 5.5 ns

t23 AHOLD, BOFF# Hold Time 1.0 ns

t24 BUSCHK#, EWBE#, HOLD, PEN# 5.0 ns

Setup Time

t25a BUSCHK#, EWBE#, PEN# Hold Time 1.0 ns

t25b HOLD Hold Time 1.5 ns

t26 A20M#, INTR, STPCLK# Setup Time 5.0 ns (4)

t27 A20M#, INTR, STPCLK# Hold Time 1.0 ns (5)

t28 INIT, FLUSH#, NMI, SMI#, IGNNE# 5.0 ns (4)

Setup Time

t29 INIT, FLUSH#, NMI, SMI#, IGNNE# 1.0 ns (5)

Hold Time

t30 INIT, FLUSH#, NMI, SMI#, IGNNE# 2.0 CLK’s (6)

Pulse Width (Async)

t31 R/S# Setup Time 5.0 ns

t32 R/S# Hold Time 1.0 ns

t33 R/S# Pulse Width (Async) 2.0 ns

t34 D0-63, DP0-7 Read Data Setup Time 3.0 ns

t35 D0-63, DP0-7 Read Data Hold Time 1.5 ns

t36 RESET Setup Time 5.0 ns

t37 RESET Hold Time 1.0 ns

t38 RESET Pulse Width, Vcc & CLK Stable 15.0 CLK’s

t39 RESET Pulse Width after Vcc & CLK 1.0 ms Power
Stable Up

t40 RESET Configuration Signals (INIT, 5.0 ns

FLUSH#) Setup Time

t41 RESET Configuration Signals (INIT, 1.0 ns

FLUSH#) Hold Time

t42a RESET Configuration Signals (INIT, 2.0 CLK’s To

FLUSH#) Setup Time (Async) RESET
Falling
Edge

t42b RESET Configuration Signals (INIT, 2.0 CLK’s To

ELECTRICAL SPECIFICATIONS 5-7

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

SYMBOL PARAMETER MIN MAX UNIT FIGURE NOTES

FLUSH#, BRDYC#, BUSCHK#) Hold RESET
Time (Async) Falling
Edge

t42c RESET Configuration Signals 3.0 CLK’s To

(BRDYC#, BUSCHK#) Setup Time RESET
(Async) Falling
Edge

t42d RESET Configuration Signals 1.0 ns To

(BRDYC#, BUSCHK#) Hold Time RESET
(Sync) Falling
Edge

t43a BF0, BF1, BF2 Setup Time 1.0 ms To

RESET
Falling
Edge

t43b BF0, BF1, BF2 Hold Time 2.0 CLK’s To

RESET
Falling
Edge

See notes after Table 5-1.

5.4 DC SPECIFICATIONS

5.4.1 Recommended Operating Conditions

Functional operation of the IDT WinChip C6 processor is
guaranteed if the conditions in Table 5-4 are met. Sustained
operation outside of the recommended operating conditions may
damage the device.

5-8 ELECTRICAL SPECIFICATIONS

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

Table 5-4. Recommended Operating Conditions

Parameter Min Max Units Notes
Operating Case Temperature 0 60 °C
VCC Voltage (3.3 V) 3.135 3.6 V
VCC Voltage (VRE) 3.45 3.6 V
VIH - High Level Input 2.0 Vcc + 0.3 V
Voltage
VIL - Low level input voltage -0.3 0.8 V
IOH - High level output 8.0 mA @ V = VOH(min)
current
(typical drive strength)
IOH - High level output 16.0 mA @ V = VOH(min)
current
(medium drive strength)
IOL - Low level output current -8.0 mA @ V = VOL(max)
(typical drive strength)
IOL - Low level output current -16.0 mA @ V = VOL(max)
(medium drive strength)

5.4.2 Maximum Ratings

While functional operation is not guaranteed beyond the
operating ranges listed in Table 5-4, the device may be subjected
to the limits specified in Table 5-5 without causing long-term
damage.
These conditions must not be imposed on the device for a
sustained period—any such sustained imposition may damage
the device. Likewise exposure to conditions in excess of the
maximum ratings may damage the device.

Table 5-5. Maximum Ratings

Parameter Min Max Units Notes
Operating Case Temperature -65 110 °C
Storage Temperature -65 150 °C
Supply Voltage (VCC) -0.5 4.0 V

ELECTRICAL SPECIFICATIONS 5-9

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

I/O Voltage -0.5 VCC + 0.5 V

or
VCC(max)

5.4.3 DC Characteristics
Table 5-4. DC Characteristics
Parameter Min Max Units Notes
VOH - High Level Output Voltage 2.4 Vcc V @ Ioh = 8mA
(typical drive strength)
VOH - High Level Output Voltage 2.4 Vcc V @ Ioh =
16mA
(medium drive strength)
VOL - Low Level Output Voltage 0 0.4 V @ Iol = -8mA
(typical drive strength)
VOL - Low Level Output Voltage 0 0.4 V @ Iol = -
16mA
(medium drive strength)
IL - Input Leakage Current ± 15 µA
ILU - Input Leakage Current for 200 µA
inputs with pull-ups
ILD - Input Leakage Current for -400 µA
inputs with pull-downs

5.4.4 Power Dissipation

Tables 5.5 and 5.6 give power consumption for the two voltage
ranges. A typical power value is not given since there is no
industry-standard definition of typical and other processor
manufacturers do not provide a public test program for their
typical values. Thus, typical values from different
manufacturers cannot be compared.
However, IDT WinChip C6 processor power consumption at 200
MHz at 3.52V while running the industry-standard Winstone 97
benchmark is 16% less than the 200 MHz 2.8V P55. Details may
be obtained from a sales representative.

5-10 ELECTRICAL SPECIFICATIONS

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

Table 5-5. Power Consumption IDT WinChip C6 @ 3.52V

Parameter Min Max Units Notes
IDD - Normal Mode
Operating Current
180 MHz 11.5 Watts
200 MHz 13.0 Watts
225 MHz 14.8 Watts
240 MHz 15.8 Watts
IDD - StopGrant / AutoHalt No
Mode Operating snooping
Current activity
180 MHz 2.8 Watts
200 MHz 4.0 Watts
225 MHz 4.4 Watts
240 MHz 4.8 Watts
IDD - StopClock Mode
Operating Current
180 MHz 3.5 Watts
200 MHz 3.5 Watts
225 MHz 3.5 Watts
240 MHz 3.5 Watts

Note:
The above power consumption is 100% tested at 70°C case and 3.52Volts.

ELECTRICAL SPECIFICATIONS 5-11

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

Table 5-6. Power Consumption IDT WinChip C6 @ 3.3V

Parameter Min Max Units Notes
IDD - Normal Mode
Operating Power
150 MHz 8.9 Watts
180 MHz 9.5 Watts
200 MHz 10.5 Watts
225 MHz 11.6 Watts
240 MHz 12.5 Watts
IDD - StopGrant / AutoHalt No
Mode Operating snooping
Power activity
150 MHz 2.0 Watts
180 MHz 2.5 Watts
200 MHz 3.3 Watts
225 MHz 4.0 Watts
240 MHz 4.6 Watts
IDD - StopClock Mode
Operating Power
150 MHz 1.4 Watts
180 MHz 1.4 Watts
200 MHz 1.4 Watts
225 MHz 1.4 Watts
240 MHz 1.4 Watts

Note:
The above power consumption is 100% tested at 70°C case and 3.3Volts.

5-12 ELECTRICAL SPECIFICATIONS

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

6 MECHANICAL SPECIFICATIONS

The IDT WinChip C6 processor packaged in a 296-pin ceramic

pin grid array (CPGA).

6.1 CPGA PACKAGE

The IDT WinChip C6 processor’s CPGA package is mechanically
compatible with Intel’s ceramic and plastic staggered pin grid
array (SPGA and PPGA) packages. See Intel’s Pentium
Processor Family Developer’s Manual, for comparison.

Figure 6-1. CPGA Pinout (Pin side View)

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37
AN AN
NC NC NC FLUSH# VCC VCC VCC VCC VCC VCC VCC VCC VCC VCC VCC A10 A6 NC VSS
AM AM
ADSC# EADS# W/R# VSS VSS VSS VSS VSS VSS VSS VSS VSS VSS VSS VSS A8 A4 A30
AL AL
VCCDET# PWT HITM# BUSCHK# BE0# BE2# BE4# BE6# SCYC NC A20 A18 A16 A14 A12 A11 A7 A3 VSS
AK AK
AP D/C# HIT# A20M# BE1# BE3# BE5# BE7# CLK RESET A19 A17 A15 A13 A9 A5 A29 A28
AJ AJ
BREQ HLDA ADS# A31 A25 VSS
AH AH
VSS LOCK# A26 A22
AG AG
VCC SMIACT# PCD A27 A24 VCC
AF AF
VSS PCHK# A21 VSS
AE AE
VCC NC APCHK# A23 NC VCC
AD AD
VSS NC INTR VSS
AC AC
VCC NC PRDY NMI R/S# VCC
AB AB
VSS HOLD SMI# VSS
AA AA
VCC NC WB/WT# INIT IGNNE# VCC
Z Z
VSS BOFF# PEN# VSS
Y Y
VCC BRDYC# NA# BF0 NC VCC
X X
VSS BRDY# BF1 VSS
W W
VCC EWBE# KEN# NC BF2 VCC
V
VSS AHOLD
IDT WinChip C6 STPCLK# VSS
V

U
VCC CACHE# INV
CPGA VCC VSS VCC
U

T
VSS M/IO#
PINOUT VCC VSS
T

S
VCC NC NC (PINSIDE VIEW) NC NC VCC
S

R R
VSS NC NC VSS
Q Q
VCC NC FERR# TRST# NC VCC
P P
VSS IERR# TMS VSS
N N
VCC D63 DP7 TDO TDI VCC
M M
VSS D62 TCK VSS
L L
VCC D61 D60 VCC NC VCC
K K
VSS D59 D0 VSS
J J
VCC D57 D58 NC D2 VCC
H H
VSS D56 NC VSS
G G
VCC D55 D53 D3 D1 VCC
F F
DP6 D51 DP5 D5 D4
E E
D54 D52 D49 D46 D42 D7 D6 VCC
D D
D50 D48 D44 D40 D39 D37 D35 D33 DP3 D30 D28 D26 D23 D19 DP1 D12 D8 DP0
C C
NC D47 D45 DP4 D38 D36 D34 D32 D31 D29 D27 D25 DP2 D24 D21 D17 D14 D10 D9
B B
NC D43 VSS VSS VSS VSS VSS VSS VSS VSS VSS VSS VSS VSS D20 D16 D13 D11
A A
NC D41 VCC VCC VCC VCC VCC VCC VCC VCC VCC VCC VCC VCC D22 D18 D15 NC

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37

MECHANICAL SPECIFICATIONS 6-1

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

Figure 6-2. CPGA Pinout (Top Side View)

37 36 35 34 33 32 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1
AN AN
VSS NC A6 A10 VCC VCC VCC VCC VCC VCC VCC VCC VCC VCC VCC FLUSH# NC NC NC
AM AM
A30 A4 A8 VSS VSS VSS VSS VSS VSS VSS VSS VSS VSS VSS VSS W/R# EADS# ADSC#
AL AL
VSS A3 A7 A11 A12 A14 A16 A18 A20 NC SCYC BE6# BE4# BE2# BE0# BUSCHK# HITM# PWT VCCDET#
AK AK
A28 A29 A5 A9 A13 A15 A17 A19 RESET CLK BE7# BE5# BE3# BE1# A20M# HIT# D/C# AP
AJ AJ
VSS A25 A31 ADS# HLDA BREQ
AH AH
A22 A26 LOCK# VSS
AG AG
VCC A24 A27 PCD SMIACT# VCC
AF AF
VSS A21 PCHK# VSS
AE AE
VCC NC A23 APCHK# NC VCC
AD AD
VSS INTR NC VSS
AC AC
VCC R/S# NMI PRDY NC VCC
AB AB
VSS SMI# HOLD VSS
AA AA
VCC IGNNE# INIT WB/WT# NC VCC
Z Z
VSS PEN# BOFF# VSS
Y Y
VCC NC BF0 NA# BRDYC# VCC
X X
VSS BF1 BRDY# VSS
W W
VCC BF2 NC KEN# EWBE# VCC
V
VSS STPCLK#
IDT WinChip C6 AHOLD VSS
V
U
VCC VSS VCC
CPGA INV CACHE# VCC
U

T
VSS VCC
PINOUT M/IO# VSS
T

S
VCC NC NC (TOP SIDE VIEW) NC NC VCC
S

R R
VSS NC NC VSS
Q q
VCC NC TRST# FERR# NC VCC
P p
VSS TMS IERR# VSS
N N
VCC TDI TDO DP7 D63 VCC
M M
VSS TCK D62 VSS
L L
VCC NC VCC D60 D61 VCC
K K
VSS D0 D59 VSS
J J
VCC D2 NC D58 D57 VCC
H H
VSS NC D56 VSS
G G
VCC D1 D3 D53 D55 VCC
F F
D4 D5 DP5 D51 DP6
E E
VCC D6 D7 D42 D46 D49 D52 D54
D D
DP0 D8 D12 DP1 D19 D23 D26 D28 D30 DP3 D33 D35 D37 D39 D40 D44 D48 D50
C C
D9 D10 D14 D17 D21 D24 DP2 D25 D27 D29 D31 D32 D34 D36 D38 DP4 D45 D47 NC
B B
D11 D13 D16 D20 VSS VSS VSS VSS VSS VSS VSS VSS VSS VSS VSS VSS D43 NC
A A
NC D15 D18 D22 VCC VCC VCC VCC VCC VCC VCC VCC VCC VCC VCC VCC D41 NC

37 36 35 34 33 32 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1

Table 6-1: CPGA Pin Cross Reference

Address Data Control Test NC Vcc Vss Reserved
Pin Pin No. Pin Pin No. Pin Name Pin No. Pin Pin No. Pin No. Pin No. Pin No. Pin No.
Name Name Name

A3 AL-35 D0 K-34 A20M# AK-08 TCK M-34 A-37 A-07 B-06 H-34

A4 AM-34 D1 G-35 ADS# AJ-05 TDI N-35 R-34 A-09 B-08 J-33

A5 AK-32 D2 J-35 ADSC# AM-02 TDO N-33 S-33 A-11 B-10 L-35

A6 AN-33 D3 G-33 AHOLD V-04 TMS P-34 S-35 A-13 B-12 Q-03

A7 AL-33 D4 F-36 AP AK-02 TRST# Q-33 W-33 A-15 B-14 Q-35

A8 AM-32 D5 F-34 APCHK# AE-05 AL-19 A-17 B-16 R-04

A9 AK-30 D6 E-35 BE0# AL-09 AN-01 A-19 B-18 S-03

A10 AN-31 D7 E-33 BE1# AK-10 AN-35 A-21 B-20 S-05

A11 AL-31 D8 D-34 BE2# AL-11 A-03 A-23 B-22 Y-35

A12 AL-29 D9 C-37 BE3# AK-12 B-02 A-25 B-24 AA-03

A13 AK-28 D10 C-35 BE4# AL-13 C-01 A-27 B-26 AC-03

A14 AL-27 D11 B-36 BE5# AK-14 AN-03 A-29 B-28 AD-04

6-2 MECHANICAL SPECIFICATIONS

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

Address Data Control Test NC Vcc Vss Reserved

Pin Pin No. Pin Pin No. Pin Name Pin No. Pin Pin No. Pin No. Pin No. Pin No. Pin No.
Name Name Name

A15 AK-26 D12 D-32 BE6# AL-15 AN-05 E-37 H-02 AE-03

A16 AL-25 D13 B-34 BE7# AK-16 H-34 G-01 H-36 AE-35

A17 AK-24 D14 C-33 BF0 Y-33 J-33 G-37 K-02

A18 AL-23 D15 A-35 BF1 X-34 L-35 J-01 K-36

A19 AK-22 D16 B-32 BF2 W-35 Q-35 J-37 M-02

A20 AL-21 D17 C-31 BOFF# Z-04 Y-35 L-01 M-36

A21 AF-34 D18 A-33 BRDY# X-04 L-33 P-02

A22 AH-36 D19 D-28 BRDYC# Y-03 L-37 P-36

A23 AE-33 D20 B-30 BREQ AJ-01 N-01 R-02

A24 AG-35 D21 C-29 BUSCHK# AL-07 N-37 R-36

A25 AJ-35 D22 A-31 CACHE# U-03 Q-01 T-02

A26 AH-34 D23 D-26 CLK AK-18 Q-37 T-36

A27 AG-33 D24 C-27 D/C# AK-04 S-01 U-35

A28 AK-36 D25 C-23 DP0 D-36 S-37 V-02

A29 AK-34 D26 D-24 DP1 D-30 T-34 V-36

A30 AM-36 D27 C-21 DP2 C-25 U-01 X-02

A31 AJ-33 D28 D-22 DP3 D-18 U-33 X-36

D29 C-19 DP4 C-07 U-37 Z-02

D30 D-20 DP5 F-06 W-01 Z-36

D31 C-17 DP6 F-02 W-37 AB-02

D32 C-15 DP7 N-05 Y-01 AB-36

D33 D-16 EADS# AM-04 Y-37 AD-02

D34 C-13 EWBE# W-03 AA-01 AD-36

D35 D-14 FERR# Q-05 AA-37 AF-02

D36 C-11 FLUSH# AN-07 AC-01 AF-36

D37 D-12 HIT# AK-06 AC-37 AH-02

D38 C-09 HITM# AL-05 AE-01 AJ-37

D39 D-10 HLDA AJ-03 AE-37 AL-37

D40 D-08 HOLD AB-04 AG-01 AM-08

D41 A-05 IERR# P-04 AG-37 AM-10

D42 E-09 IGNNE# AA-35 AN-09 AM-12

D43 B-04 INIT AA-33 AN-11 AM-14

D44 D-06 INTR AD-34 AN-13 AM-16

D45 C-05 INV U-05 AN-15 AM-18

D46 E-07 KEN# W-05 AN-17 AM-20

D47 C-03 LOCK# AH-04 AN-19 AM-22

D48 D-04 M/IO# T-04 AN-21 AM-24

D49 E-05 NA# Y-05 AN-23 AM-26

MECHANICAL SPECIFICATIONS 6-3

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

Address Data Control Test NC Vcc Vss Reserved

Pin Pin No. Pin Pin No. Pin Name Pin No. Pin Pin No. Pin No. Pin No. Pin No. Pin No.
Name Name Name

D50 D-02 NMI AC-33 AN-25 AM-28

D51 F-04 PCD AG-05 AN-27 AM-30

D52 E-03 PCHK# AF-04 AN-29 AN-37

D53 G-05 PEN# Z-34

D54 E-01 PRDY AC-05

D55 G-03 PWT AL-03

D56 H-04 RESET AK-20

D57 J-03 R/S# AC-35

D58 J-05 SCYC AL-17

D59 K-04 SMI# AB-34

D60 L-05 SMIACT# AG-03

D61 L-03 STPCLK# V-34

D62 M-04 W/R# AM-06

D63 N-03 WB/WT# AA-05

VCC2DET# AL01

6-4 MECHANICAL SPECIFICATIONS

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

Figure 6-3. CPGA Package Dimensions

MECHANICAL SPECIFICATIONS 6-5

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

Figure 6-4. CPGA Dimensions

Millimeters Inches
Symbol Min Max Notes Min Max Notes
A1 0.69 0.84 Lid 0.027 0.033 Lid
A2 3.31 3.81 Lid 0.130 0.150 Lid
B 0.43 0.51 0.017 0.020
D 49.28 49.78 1.940 1.960
D1 45.59 45.85 1.795 1.805
E1 2.29 2.79 0.090 0.110
L 3.05 3.30 0.120 0.130
N 296 Lead 296 Lead
Count Count
S1 1.52 2.54 0.060 0.100

6-6 MECHANICAL SPECIFICATIONS

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

7 THERMAL SPECIFICATIONS

7.1 INTRODUCTION
The IDT WinChip C6 is specified for operation with device case
temperatures in the range of 0°C to 70°C. Operation outside of
this range will result in functional failures and potentially
damage the device.
Care must be taken to ensure that the case temperature
remains within the specified range at all times during
operation. An effective heat sink with adequate airflow is
therefore a requirement during operation.

7.2 TYPICAL ENVIRONMENTS

Typical thermal solutions involve three components: a heatsink,
an interface material between the heatsink and the package,
and a source of airflow. The best thermal solutions rely on the
use of all three components. To the extent that any of these
components are not used, the other components must be
improved to compensate for such omission. In particular, the
use of interface material such as thermal grease, silicone paste,
or graphite impregnated paper can make a 40°C difference in
the case temperature (see Table 7-4). Likewise, the imposition
of airflow is realistically a requirement (see Table 7-1).

7.3 MEASURING TC
The Intel Pentium Processor Developer’s Manual describes
proper thermal measuring techniques in detail in Chapter 10.
The case temperature (TC) should be measured by attaching a
thermocouple to the center of the IDT WinChip C6 package. The
heat produced by the processor is very localized so measuring
the case temperature anywhere else will underestimate the case
temperature.
The presence of a thermocouple is inherently invasive; effort
must be taken to minimize the effect of the measurement. The
thermocouple should be attached to the processor through a
small hole drilled in the heatsink. Thermal grease should be
used to ensure that the thermocouple makes good contact with
the package, but the thermocouple should not come in direct
contact with the heatsink.

THERMAL SPECIFICATIONS 7-1

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

Physical Test Conditions

Case temperature measurements should be made in the worst
case operating environments. Ideally, systems should be
maximally configured, and tested at the worst-case ambient
temperature.

Test Patterns
During normal operation the processor attempts to minimize
power consumption. Consequently, normal power consumption
is much lower than the maximum power consumption. Thermal
testing should be done while running software which causes the
processor to operate at its thermal limits. Your IDT sales
representative can supply you with an executable program
which will maximize power consumption.

7.4 ESTIMATING TC
The IDT WinChip C6 processor’s case temperature can be
estimated based on the general characteristics of the thermal
environment. This estimate is not intended as a replacement
for actual measurement.
Case temperature can be estimated from Tables 7-1 and 7-2
below, where,
TA ≡ Ambient Temperature
TC ≡ Case Temperature
θCA ≡ case-to-ambient thermal resistance
θJA ≡ junction-to-ambient thermal resistance
θJC ≡ junction-to-case thermal resistance
P ≡ power consumption (Watts)
and,
TJ = TC + (P * θJC)
TA = TJ – (P * θJA)
TA = TC – (P * θCA)
θCA = θJA – θJC

7-2 THERMAL SPECIFICATIONS

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

Table 7-1. CPGA θ JC and θ JA

θ JA (°°C/Watt) vs. Laminar Airflow
(linear ft/min)
Heat Sink in θ JC (°°C/Watt) 0 100 200 400 600 800
Inches
(height)
0.25 2.0 10.3 9.2 7.8 5.7 4.8 4.2
0.35 2.0 10.0 8.7 7.2 5.2 4.5 4.0
0.45 2.0 8.6 8.2 6.5 4.8 4.1 3.7
0.55 2.0 9.3 7.7 5.9 4.4 4.8 3.5
0.65 2.0 8.9 7.2 5.5 4.2 3.6 3.3
0.80 2.0 8.2 6.5 5.1 4.0 3.4 3.2
1.00 2.0 7.5 5.9 4.8 3.8 3.3 3.0
1.20 2.0 7.1 5.5 4.5 3.6 3.2 3.0
1.40 2.0 6.6 5.1 4.2 3.4 3.1 2.9
No Heat Sink 2.5 15.6 14.3 12.9 10.0 8.6 7.7

Environment: these estimates assume the use of thermal grease between the processor and the
heatsink. Heatsinks are 1.95” square.

Table 7-2. IDT WinChip C6 Power Consumption

Frequency (MHz) VCC (Volts) Max Power (Watts)
180 3.3 9.5
3.52 11.5
200 3.3 10.5
3.52 13
225 3.3 -
3.52 14.8
240 3.3 -
3.52 15.8

Example:
For a system with:
• 200 MHz IDT WinChip C6 @ 3.52V
• TA = 35°C, (ambient air around the heatsink)

THERMAL SPECIFICATIONS 7-3

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

• 0.65” Heat Sink w/ 400 fpm laminar air flow

The case temperature can be calculated as
TC = TA + (P * θCA)
TC = TA + (P * (θJA - θJC))
TC = 35 + (13 * (4.2-2.0))
TC = 63.6 °C

7.5 RECOMMENDED THERMAL SOLUTIONS

Table 7-3 below offers several off-the-shelf thermal solutions
which are known to provide adequate cooling for the IDT
WinChip C6. The table indicates the maximum chassis internal
ambient temperature that the heatsink can handle while still
keeping the processor’s case under 70°C These kits were
measured with silicone paste as an attach media except as
noted.
For reference, Table 7-4 shows the effectiveness of the Acadia
Technology CPB-15502-02 heatsink/fan with different attach
media.
Table 7-3. Heatsink / Fan Kits
Manufacturer Part Max Chassis TA
Number
180/3.3V 180/3.52V 200/3.3V 200/3.52V

Aavid1 26437 51.0°C 47.6°C 48.3°C 43.1°C

Acadia CPB- 56.3°C 52.7°C 53.0°C 49.0°C
Technology 15502-02
Cooler Master TP5-5015 53.0°C 49.2°C 51.0°C 46.0°C
Cooler Master TP5-5020 52.3°C 49.0°C 51.1°C 46.0°C

1
The Aavid 26437 has an attached Graphite sheet. This eliminates the need for
thermal grease or epoxy.

7-4 THERMAL SPECIFICATIONS

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

Table 7-4. Interface Material Effectiveness (CPB-15502-02)

Manufacturer Part Max Chassis TA Type
Number (180/3.3V)
Omega OT-201 56.3°C Silicone Paste
- - 45.0°C Graphite Paper
- - 13.6°C Nothing

7.6 CONTACTS
Table 7-5. Heatsink / Fan Kit Contacts
Manufacturer Contact Phone # Address
Aavid Chris Chapman (603)528-3400 One Kool Path

PO Box 400

Laconia, NH 03247

Acadia Technology Chung Bao (408)747-1349 1010 Morse Ave.

Suite 4

Sunnyvale, CA 94089

Cooler Master Jerry Chen (510)770-0149 115 Fourier Ave.

Fremont, CA 94539

Table 7-6. Interface Material Contacts

Manufacturer Contact Phone # Address
Omega Technologies (203)359-1660 One Omega Drive

Stamford, CT 06907

THERMAL SPECIFICATIONS 7-5

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

APPENDIX A. MACHINE SPECIFIC REGISTERS

A.1 GENERAL
Tables A-1 and A-2 summarize the IDT WinChip C6 processor
machine-specific registers (MSRs). Further description of each
MSR follows the table. MSRs are read using the RDMSR
instruction and written using the WRMSR instruction.
There are four basic groups of MSRs (not necessarily with
contiguous addresses). Other than as defined below, a reference
to an undefined MSR causes a General Protection exception.
1. Those that are very similar in function (but possibly
different in some detail) to the Pentium processor MSRs.
Generally, the same MSR address is used. These
registers can have some utility to low-level programs
(like BIOS).
Note that some of the first sixteen Pentium MSRs
(addresses 0 to 15) have no function in the IDT WinChip
C6 processor. These MSRs do not cause a GP when used
on the IDT WinChip C6 processor; instead, reads to these
MSRs return zero, and writes are ignored.
2. Memory Configuration Registers which use MSR
addresses that are not used on the Pentium processor.
These MSRs define memory ranges with associated
attributes. These MSRs are similar to the Pentium Pro
processor MTRRs and to the Cyrix 6x86MX processor’s
ARR registers.
3. MSRs used for cache and TLB testing. These use MSR
addresses that are not used on the Pentium. These test
functions are very low-level and complicated to use. They
are not documented in this datasheet but the information
will be provided to customers given an appropriate
justification.
4. There are some undocumented internal-use MSRs used
for low-level hardware testing purposes. Attempts to read
or write these undocumented MSRs cause unpredictable
and disastrous results; so don’t use MSRs that are not
documented in this datasheet!

MACHINE SPECIFIC REGISTERS A-1

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

MSRs are not reinitialized by the bus INIT interrupt; the setting
of MSRs is preserved across INIT.

Table A-1. Category 1 MSRs (Functionally Similar to Pentium)

MSR MSR Name ECX EDX EAX Type Notes
no MSR 00h-01h n/a n/a RW 1
TR1 Test Register 1 02h n/a Control bits RW 3
no MSR 03h-0Dh n/a n/a RW 1
TR12 Test Register 12 0Eh n/a Control bits RW 3
no MSR 0Fh n/a n/a RW 1
TSC Time Stamp Counter 10h Count[63:32] Count[31:0] RW 2
EC_CTRL Event Counter Control 11h n/a Control bits RW 2
EC0 Event Counter 0 12h Count[39:32] Count[31:0] RW 2
EC1 Event Counter 1 13h Count[39:32] Count[31:0] RW 2
FCR Feature Control Reg 107h n/a FCR value RW 4
FCR2 Feature Control Reg 2 108h FCR2_Hi FCR2 value RW 5
FCR3 Feature Control Reg 3 109h FCR3_Hi FCR3 value WO 5

Notes
1. Pentium processors have MSRs at these addresses. On the
IDT WinChip C6 processor, reads to these addresses return
zero and writes are ignored.

2. Functionally similar to the same Pentium MSR. However,

some minor details are different.

3. A subset of the same Pentium MSR—only those bits

meaningful to the IDT WinChip C6 processor have any effect;
the rest read as 0 and are ignored when written.

4. Conceptually similar to the Pentium MSR 0Eh (“TR12”) that

controls detailed functions like disabling the caches. The IDT
WinChip C6 processor controls are different, thus the FCR is
placed at a different address than the Pentium TR12 register.

5. FCR2 and FCR3 provide system software with the ability to

specify the Vendor ID string returned by the CPUID
instruction .

A-2 MACHINE SPECIFIC REGISTERS

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

Table A-2. Category 2 MSRs (Memory Configuration Registers)

MSR ECX EDX EAX Type Notes
MCR 0 110h Base Address Address Mask WO
[31:12] [31:12]
& Ctrl Value
[11:0]
MCR 1 111h same as above same as above WO
MCR 2 112h same as above same as above WO
MCR 3 113h same as above same as above WO
MCR 4 114h same as above same as above WO
MCR 5 115h same as above same as above WO
MCR 6 116h same as above same as above WO
MCR 7 117h same as above same as above WO
MCR_CTRL 120h - control value WO

MACHINE SPECIFIC REGISTERS A-3

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

A.2 CATEGORY 1 MSRS

02h: TR1 (Pentium Processor Parity Reversal Register)

31:2 1 0
Reserved (Ignored on write; returns 0 on read) NS Res
30 1 1

Both the IDT WinChip C6 processor and the Intel Pentium

processor have a MSR 02 bit 1 that performs the same function
on an IDT WinChip C6 processor as on a Pentium processor.
Other bits return 0 when read and are ignored when written.
NS: 0 = Assert #IERR and cause Shutdown on internal
parity error
1 = Assert #IERR and do not cause Shutdown on
internal parity error

0Eh: TR12 (Pentium Processor Feature Control)

31:10 9 8:7 6 5:4 3 2:0
Reserved ITR Res AHD Res CI Res
(Ignored on write;
returns 0 on read)
22 1 2 1 2 1 3

Both the IDT WinChip C6 processor and Pentium processor

have MSR 0E bits 3, 6, and 9 that perform the same functions
on an IDT WinChip C6 processor as on a Pentium processor.
Other bits return 0 when read and are ignored when written.
CI: 0 = Ignored.
1 = Same as for the Pentium processor: Cache line
fills (to both caches) are suppressed; all cache misses
are performed as single transfer cycles. The PCD
output pin is not affected. Note that the caches are
not flushed.
AHD: 0 = Ignored.
1 = Same as for the Pentium processor: disable
Autohalt Powerdown function

A-4 MACHINE SPECIFIC REGISTERS

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

ITR: 0 = Ignored.
1 = Same as for the Pentium processor: enable SMM
I/O Restart function

10h: TSC (Time Stamp Counter)

Both the IDT WinChip C6 processor and the Pentium processor
have a 64-bit MSR that materializes the Time Stamp Counter
(TSC). Both systems increment the TSC once per processor
clock.

On the IDT WinChip C6 processor, the MSR (and the value

returned by the RDTSC instruction) is an alias for the internal
event-counter MSRs (CTR0/CTR1). In normal system operation,
the TSC register counts internal processor clocks.

However, if the user code changes the item that CTR0 or CTR1
is counting (see the counter MSR descriptions), then the TSC
register also changes what it is counting. There is no practical
reason why the machine-specific event counting should be
changed by software. On a Pentium processor, the TSC is a
separate counter from CTR0/CTR1.

11h: CESR (Control & Event Select Register)

31:24 23:16 15:8 7:0
Reserved CTR1 Control Reserved CTR0 Control
8 8 8 8
Both the IDT WinChip C6 processor and Pentium have an MSR
that contains bits defining the behavior of the two hardware
event counters: CTR0 and CTR1.
The CTR0 and CTR1 control fields define which of several
possible events can be counted for each counter. Each counter
has the same set of possible events.
The events that can be counted, and their identification
numbers, are different from the Pentium processor events
(which are different from the Pentium Pro processor events).
The Pentium processor has only six bits to identify the event
counter, but has additional controls (such as event versus clock
counting) in bits 9-6 of each control field.

MACHINE SPECIFIC REGISTERS A-5

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

The CESR should be written before the associated CTR0 and

CTR1 are written to initialize the counters. The counts are not
necessarily perfectly exact; the counters are intended for use
over a large number of events and may differ by one or two
counts from what might be expected.
Most counter events are internal implementation-dependent
debug functions having no meaning to software. The counters
that can have end-user utility are:

Event Description
0 Internal clocks (default event for CTR0)
1 valid cycles reaching writebacks
2 X86 instructions
71 data read cache misses
74 data write cache misses
99 instruction fetch cache miss

12h-13h: CTR0 & CTR1 (Event Counters 0 & 1)

Both the IDT WinChip C6 processor and Pentium processor
have two 40-bit hardware event counters (bits 31:8 of EDX are
ignored).

107h: FCR (Feature Control Register)

The FCR controls the major optional feature capabilities of the
IDT WinChip C6 processor. It is analogous to the Pentium
processor TR12 (actually MSR 0Eh) that controls things like
BTB enable, cache enable, and so forth. The Cyrix 6x86MX
processor’s CCRs (Configuration Control Registers) perform a
similar function, as does the AMD-K6 processor’s HDCR MSR.
Table A-2 contains the bit values for the FCR. The defaults
settings shown for the FCR bits are not necessarily exact. The
actual settings can be changed as part of the manufacturing
process and thus a particular IDT WinChip C6 processor version
can have slightly different default settings than shown here. All
reserved bit values of the FCR must be preserved by using a
read-modify-write sequence to update the FCR.

A-6 MACHINE SPECIFIC REGISTERS

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

Table A-2. FCR Bit Assignments

Bit Name Description Default
0 Reserved 0
1 ECX8 Enables CPUID reporting CX8 0
2 EIERRINT Enables INT18 (Machine Check) for internal 0
errors
3 DPM Disable dynamic power management 0
4 DMCE Disables Machine Check Exception 0
5 DSTPCLK Disables supporting STPCLK 0
6 ELINEAR Enables Linear Burst Mode 0
7 DSMC Disables strict cache coherency (self-modifying- 0
code)
8 EDCTLB Enables D-Cache for updates to PDE/PTE 0
9 EMMX Enables MMX-compatible instructions. 1
10 Reserved 0
11 DPDC Disables Page Directory cache 0
12 Reserved 0
13 DIC Disables I-Cache. 0
14 DDC Disables D-Cache. 0
15 DNA Disables bus pipelining (NA response) 0
16 ERETSTK Enables CALL-RET Stack operation 1
17 Reserved 0
18 Reserved 0
19 Reserved 0
20 Reserved 0
21 Reserved 0
22:25 SID Stepping ID 0
26 Reserved 0
27 Reserved 0
28 ADIVFLG Selects alternate EFLAGS results for divide 0
29 DCPUID Disables CPUID instruction 0
30 EMOVTR Enables move-to-test register instructions 0
31 Reserved 0

MACHINE SPECIFIC REGISTERS A-7

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

ECX8: 0 = The CPUID instruction does not report the

presence of the CMPXCHG8B instruction (CX8 =
0). The instruction actually exists and operates
correctly, however.
1 = The CPUID instruction reports that the
CMPXCHG8B instruction is supported (CX8 = 1).
EIERRINT: 0 = Normal internal error behavior (IERR and
possible Shutdown).
1 = Causes INT18 instead of Shutdown for internal
error.
DPM: 0 = Normal dynamic power management behavior.
1 = Disables all dynamic power management.
DMCE: 0 = Machine Check exception enabled.
1 = Disable Machine Check exception; a bus check
or internal error condition does not cause an
exception.
DSTPCLK: 0 = STPCLK interrupt properly supported.
1 = Ignores SPCLK interrupt.
ELINEAR: 0 = Interleaved burst ordering is enabled.
1 = Linear burst ordering is enabled.
DSMC: 0 = Strict cache coherency is enabled to support
Pentium processor style self-modifying code.
1 = Disables strict cache coherency. I-cache/D-
cache coherent only if branch is taken after store
instruction which modifies instructions to be
executed subsequently.
EDCTLB: 0 = Updates to the accessed and dirty bits in
PDE/PTE entries are performed using the locked
read-modify-write semantics which flushes the
data from the D-Cache (like Pentium processor).
1 = Enables D-Cache for updates to accessed and
dirty bits in PDE/PTE.
EMMX: 0 = Disables MMX-compatible instructions: they
decode as invalid instructions.
1 = Enables MMX-compatible instructions.
DPDC: 0 = Enables use of internal Page Directory Cache.
1 = Disables use of internal Page Directory Cache.

A-8 MACHINE SPECIFIC REGISTERS

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

DIC: 0 = Enables use of I-Cache.

1 = Disables use of I-Cache: cache misses are
performed as single transfer bus cycles, PCD is de-
asserted. This overrides any setting of CR0.CD and
CR0.NW.
DDC: 0 = Enables use of D-Cache.
1 = Disables use of D-Cache: same semantics as for
DIC except for D-Cache.
DNA: 0 = Enables bus pipelining operation.
1 = Disables bus pipelining operation: bus signal
NA is ignored.
ERETSTK: 0 = Disables CALL-RETurn stack function.
1 = Enables CALL-RETurn stack function: RET
branch target prediction is performed.
ADIVFLG: 0 = The EFLAGS setting after integer divide
instructions is unique to the IDT WinChip C6
processor. These flag settings are undefined in the
x86 architecture.
1 = The EFLAGS setting after integer divide of 5 by
2 is the same as on a Pentium processor (this is
used by some old code to identify the processor)
DCPUID: 0 = The CPUID instruction is supported.
1 = The CPUID instruction is disabled and causes
an invalid instruction exception.
EMOVTR: 0 = The Intel486 move-to/from-test register
instructions are not supported and their behavior is
the same as on a Pentium processor (invalid
instruction exception).
1 = The test register instructions do not cause an
invalid instruction exception but rather are treated
as NOPs.

MACHINE SPECIFIC REGISTERS A-9

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

108h: FCR2 (Feature Control Register 2)

This MSR contains more feature control bits — many of which
are undefined. It is important that all reserved bits are
preserved by using a read-modify-write sequence to update the
MSR.
63:32
Last 4 characters of Alternate Vendor ID string

31:15 14 13:12 11:8 7:4 3:0

Reserved AVS Res Family Model Res
ID ID
17 1 2 4 4 4
AVS: 0 = The CPUID instruction vendor ID is
“CentaurHauls”
1 = The CPUID instruction returns the alternate
Vendor ID. The first 8 characters of the alternate
Vendor ID are stored in FCR3 and the last 4
characters in FCR2[63:32]. These 12 characters
are undefined after RESET and may be loaded by
system software using WRMSR.
Family ID: This field will be returned as the family ID field by
subsequent uses of the CPUID instruction
Model ID: This field will be returned as the model ID field by
subsequent uses of the CPUID instruction

109h: FCR3 (Feature Control Register 3)

This MSR contains the first 8 characters of the alternate Vendor
ID. The alternate Vendor ID is returned by the CPUID
instruction when FCR2[AVS] is set to 1. FCR3 is a write-only
MSR.
63:32
First 4 characters of Alternate Vendor ID string

31:0
Middle 4 characters of Alternate Vendor ID string

A-10 MACHINE SPECIFIC REGISTERS

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

A.3 MEMORY CONFIGURATION REGISTERS

General
The IDT WinChip C6 processor provides extensions over the P55
to define variable size memory ranges with associated special
attributes. These Memory Configuration Registers (MCRs) are
similar to the MTRRs of the Pentium Pro processor and the
Address Region Registers (ARRs) of the Cyrix 6x86MX
processor. All of these approaches perform similar functions but
differ in specifics.
The basic function performed is to define memory regions and
special attributes for these regions such as byte-combining and
weakly ordered stores. These special attributes are deviations
from formal x86 architectural behavior but, in practice, work
fine for specific memory regions in a PC architecture. The
advantage of these special attributes is improved performance
for the affected memory regions.

Memory Configuration Registers

The IDT WinChip C6 processor has eight MCRs, each appearing
as a 64-bit MSR. The default value at reset is zero for all fields.
This value causes all memory to have normal x86 attributes of
no byte-combining and strongly ordered writes.
Note that the MCRs are write-only.
The MCR format is:
EDX 31:12 EDX 11:0 EAX 31:12 EAX 11:5 EAX 4:0
Base Address of Res Mask Defined Region Res Attributes
Region
20 12 20 7 5

Base
This is the starting physical address of the memory region. Each
definable memory region starts at a 4-KB page boundary; thus
the low order 12 bits of the address are ignored.

MACHINE SPECIFIC REGISTERS A-11

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

Mask
This is a bit mask defining the size of the memory region. A
memory region hit exists for a MCR if:
Mask address AND memory address = Base address AND
mask address in all bit positions (31:12)

Example 1.
For example, consider the memory range from 0x000A0000 to
0x000BFFFF. This is most efficiently done by masking off the
low order bits that constitute the range.
Viewing the addresses in binary:
0x000A0000 = 0000 0000 0000 1010 0000 0000 0000 0000
0x000BFFFF = 0000 0000 0000 1011 1111 1111 1111 1111
Notice that the upper 15 bits are identical, whereas the lower
17 bits define the address within the range. So a single MRD
can describe this range as:
MCR Base = 0000 0000 0000 1010 0000 or 0x000A0
MCR Mask = 1111 1111 1111 1110 0000 or 0xFFFE0
Note that the lower twelve bits of the mask and base are
ignored in the match calculation.

Example 2.
Consider a more complex scenario where a range from
0x00080000 to 0x00017FFF is required. Shown in binary as:
0x00080000 = 0000 0000 0000 1000 0000 0000 0000 0000
0x0017FFFF = 0000 0000 0001 0111 1111 1111 1111 1111
In this case, the upper 11 bits are identical throughout the
range. The next two bits vary, but not all combinations are
part of the desired range. The range has to be broken down
into two ranges that have common base bits where all
combinations of the lower order bits are within the original
region. This implies:
0x00080000 = 0000 0000 0000 1000 0000 0000 0000 0000
0x000FFFFF = 0000 0000 0000 1111 1111 1111 1111 1111
and
0x00100000 = 0000 0000 0001 0000 0000 0000 0000 0000
0x0017FFFF = 0000 0000 0001 0111 1111 1111 1111 1111

A-12 MACHINE SPECIFIC REGISTERS

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

These map into MCRs:

MCR0 Base = 0000 0000 0000 1000 0000 or 0x00080
MCR0 Mask = 1111 1111 1111 1000 0000 or 0xFFF80
and
MCR1 Base = 0000 0000 0001 0000 0000 or 0x00100
MCR1 Mask = 1111 1111 1111 1000 0000 or 0xFFF80

Attributes
There are five bits of attribute control defined for each memory
region as described in Table A.6-1
Table A.6-1
Bit Description Default Notes
0 Enables store combining on non-stack non-string 0
writes.
1 Enables store combining on string instruction writes. 0
2 Enables store combining on stack instruction writes. 0
4:3 Defines write-ordering strategy for memory region. 00
00: All writes strongly ordering
01: String writes weakly ordered
10: Stack writes weakly ordered
11: All writes weakly ordered

MACHINE SPECIFIC REGISTERS A-13

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

MCR Control Register

The MCR_CTRL MSR controls various pervasive behaviors of
byte-combining and write-ordering. The store combining
definitions for string and non-stack and non-string are defined
in the following table.
Table A.7-1
Bit Description Default Notes
1:0 Store combining definition for non-stack & non-string. 00
00: Forward Combining
01: Forward and Overlapped Combining
10: Forward and Reverse Combining
11: Forward, Reverse, and Overlapped Combining
3:2 Store combining definition for string. 00
00: Forward Combining
01: Forward and Overlapped Combining
10: Forward and Reverse Combining
11: Forward, Reverse, and Overlapped Combining
4 Weak Write-Ordering Enable 0
5:19 Reserved 0
24:20 Reserved 11111

Store-Combining
The IDT WinChip C6 processor’s store-combining feature allows
multiple stores into be combined into a single bus store. This is
permissible if the stores are destined to the same 8-byte memory
address.
Store-combining can greatly reduce the memory bandwidth
requirements of writes that miss the cache. However, if the
associated writes are destined to memory mapped I/O locations,
problems can arise.
For example, if an ISA bus 8-bit device is controlled with a data
register at address 0 and a control register at address 1, and the
control register must be written before the data can be written,
it is possible for the order of writes to be changed if byte-
combining is inappropriately configured.
If, for example, the device writes to address 1 and then to
address 0 and the two are combined, a 16-bit word will go to the
ISA bus where it will be split for the 8-bit device into two writes.
Unfortunately, most ISA bridges split 16-bit operands into two
transfers with the low byte first. Consequently, the order of the
two writes are reversed.

A-14 MACHINE SPECIFIC REGISTERS

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

To eliminate this problem, the IDT WinChip C6 processor is

very configurable. Aside from disabling byte-combining, it is
also possible to limit the type of instructions that are allowed to
combine. Further, it is possible to prevent the processor from
combining in reverse address order. Lastly, it is possible to
prevent the processor from combine-matching (overlapping) byte
addresses.

Weak Write-Ordering
The Pentium processor normally mandates that writes occur in
the memory hierarchy in the same order as they occur in the
code execution. This is termed strong write-ordering. This
restriction can be a performance impact in that it blocks
processor execution when a store hits an E or M line in the
cache if another store is waiting to be retired on the bus.
Normally systems do not require strong write-ordering unless
they have bus-mastering I/O devices that use memory mapped
I/O for control purposes. (Most DMA devices are slaves, and do
not use memory-mapped I/O. The floppy controller, for example,
is a DMA device, but does not use memory-mapped I/O.)
However, since there are devices that could not perform
correctly with weak write ordering, this function should only be
used in systems where the type of peripherals are tightly
controlled and known to not require strongly ordered writes.
Weak write ordering should never be turned on by a generic
BIOS, for example.

Combining Ranges
It is possible to describe fairly complex ranges with a few
descriptors. Generally, this does not involve overlapping MCRs.
However, overlapping ranges are permitted and their behavior
is useful in some cases.
The behavior of an access to a given memory location is defined
by the logical OR of the attribute bits of the MCRs it matches.
So, if a memory location does not match any MCRs, its
aggregate attribute is 0. If, on the other hand, it matches two
MCRs, one with an attribute of 0x18 and the other with an
attribute of 0x02, the aggregate attribute is 0x1A. This enables
weak writes on all accesses and allows store-combining on stack
accesses.

MACHINE SPECIFIC REGISTERS A-15

 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

APPENDIX B. COMPATIBILITY

B.1 INTRODUCTION
In general, the IDT WinChip C6 processor is exactly compatible
with both the bus and software-visible architecture of the Intel
Pentium processor.
An IDT WinChip C6 processor can plug into existing Intel
Pentium-based PC system boards and operate without requiring
change to the system hardware. Also, an IDT WinChip C6
processor can run all existing industry-standard PC object-code
operating systems and application programs.
However, all processors developed for use in PCs (“x86”
processors) have some minor incompatibilities in low-level
implementation-dependent functions. For example, it is possible
to write esoteric software for the Intel 486 processor (cache tests,
for example) that produces different results when run on the
supposedly compatible Intel Pentium processor.
Similarly, there are low-level incompatibilities between the Intel
Pentium and the Intel Pentium Pro processors. Similarly, there
are low-level incompatibilities among all x86 “clone” processors
such as the AMD-K6 and the Cyrix 6x86MX processors. The IDT
WinChip C6 processor has similar low-level differences with the
various Intel and x86 clone processors.
Fortunately, these technical incompatibilities among x86
implementations are in areas that have no meaningful use to
most programs, and that are well-understood by software
developers (and are thus avoided). Therefore, in practice, these
types of differences pose no real barriers to program
compatibility across various implementations.
This appendix summarizes areas where the IDT WinChip C6
processor differs in behavior from the Intel Pentium processor.
These differences are generally “don’t cares”; that is, they are
transparent to system hardware and programs.

COMPATIBILITY B-1
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

B.2 BUS COMPATIBILITY

Bus Cycle Activity

When compared cycle by cycle, the IDT WinChip C6 and Intel
Pentium processors do not have exactly the same bus cycles.
This is anticipated, unavoidable, and desirable (the IDT
WinChip C6 processor provides increased bus performance).
This difference results from the IDT WinChip C6 processor’s
different internal architecture and larger cache. This is not an
issue because the Intel Pentium processor itself varies at
different frequencies and all other competitive processors also
have differing bus cycle activity.

Bus Alignment
Although the Intel Pentium processor has a 64-bit bus, it splits
loads and stores that cross 32-bit boundaries. The IDT WinChip
C6 processor splits memory loads and stores at 64-bit
boundaries. However, I/O reads and writes are split on 32-bit
boundaries.
Like the Pentium processor, when split cycles are required, the
IDT WinChip C6 processor performs the higher address’ access
first, then performs the lower address’ access.
The IDT WinChip C6’s bus alignment is not anticipated to be a
compatibility issue. The Cyrix 6x86 and AMD-K5 processors
also split memory loads and stores at 64-bit boundaries. (Also,
the Cyrix 6x86 performs the low addressed access first except for
32-bit misaligned I/O’s in which case the higher access is
performed first.)

Bus Data with Inactive Byte Enables

For each of the 8 bytes on the data bus there is a byte enable
signal that indicates whether the corresponding byte is part of
the transaction. Bits on the data bus with inactive byte enables
have undefined values. During I/O writes or stores to memory
P54 drives 0x00 or a mirror of the corresponding byte in the
other double word on the bus. The AMD-K6 drives 0xFF or a
mirror of the corresponding byte in the other double word on the
bus. The IDT WinChip C6 (except stepping 0) is compatible
with P54 during an I/O write bus cycle. On a memory write the
IDT WinChip C6 drives undefined values on the data bus for
bytes that are not enabled, a behavior that is different from
either P54 or the AMD-K6. This difference on memory writes
has not caused compatibility problems.

B-2 COMPATIBILITY
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

Snoop Responsibility Pickup

The Pentium processor assumes responsibility for incoming
cache lines at the point where it determines the cacheability of
the line. The cacheability is determined at the first assertion of
NA# or BRDY# for the cycle. Subsequent to cacheability
determination, the processor will respond to snoops for that line
even though it is not completely read into the processor.
The IDT WinChip C6 processor generally mimics this behavior.
However, if a bus cycle is pipe-lined over a previous cycle and
gets a BOFF# assertion before its cacheability for the second
line has been determined, the IDT WinChip C6 processor
assumes responsibility for subsequent snoops to both lines when
cacheability is determined for the first retried line.

Descriptor Updates
The exact size of the locked bus transactions used to update the
accessed bit in non-accessed descriptors is slightly different
between the Pentium processor and the IDT WinChip C6
processor.

8-Byte Writes
Eight-byte aligned locked 8-byte read-modify-write sequences (a
locked CMPXCHG8B instruction) are performed as one locked 8-
byte read followed by two locked 4-byte writes. The Pentium
processor performs the write as an 8-byte locked write.

TLB Retries
Intel processors are not consistent about how and when they
take page protection exceptions. The Pentium Pro’s operation is
architecturally cleaner than the Pentium’s operation, so the IDT
WinChip C6 mimics the Pentium Pro’s behavior. If a memory
address hits either of the TLB’s and the associated TLB entry
would indicate that a protection page fault should be taken, the
page tables are retried to ensure that the TLB entry is up to
date. Only if the retried TLB entry still indicates that a page
protection exception should occur does the exception actually
take place.

Table Walk D and A Bit Updates

The IDT WinChip C6 processor does not snoop its instruction
cache during table walks. Consequently, if the access (A) or
dirty (D) bit need to be updated, and the table or directory entry
is in the L1 instruction cache, the L1 instruction cache does not
reflect the updated value.

COMPATIBILITY B-3
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

While it is possible to create code that detects this inconsistency,

it is highly unlikely that any application or operating system
relies on the update.

SCYC
The SCYC signal is loosely defined on the Intel Pentium
processor. On the IDT WinChip C6 processor, SCYC is asserted
only during external locked read-modify-write cycles that are
unaligned.

STPCLK# in Auto Halt State

If the processor sees an assertion on STPCLK# while resting in
the Auto Halt Power Down State, it awakens briefly, and goes
into the Stop Grant State. This differs from the Intel Pentium
processor, which ignores STPCLK# in Auto Halt. The AMD-K6
x86 processor behaves the same as the IDT WinChip C6
processor.

B.3 INTEGER INSTRUCTION COMPATIBILITY

CPUID Vendor ID String
The vendor identification string returned by the CPUID
instruction is different, of course, among the various Pentium-
compatible processor manufacturers. A very few programs
(games) are dependent upon the “GenuineIntel” string returned
by the Intel processors and thus will not perform correctly with
the default “CentaurHauls” string returned by the IDT WinChip
C6 processor.
The IDT WinChip C6 processor has a feature that allows the
user to define the vendor ID string returned by the CPUID
instruction. This allows programs that are dependent on specific
vendor ID strings to be run. This feature is described in
Appendix A.

CPUID Feature Flags

As defined in Chapter 3, the feature definition flags returned by
the CPUID instruction differ among various Pentium-
compatible processors.

B-4 COMPATIBILITY
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

Machine-Specific Registers
As defined in Appendix A, the machines-specific registers
(MSR’s) of a IDT WinChip C6 processor are just that: machine-
specific. That is, there are differences between the IDT WinChip
C6 processor MSR’s and the Pentium processor’s MSR’s (and all
other x86 processors).

Undefined EFLAGS Settings

The Intel Pentium documentation defines the setting of the
EFLAGS bits for several instructions as “undefined”. In some
cases the setting of undefined flags is consistent among
Pentium-compatible processors and some programs depend upon
these settings. In these cases, the IDT WinChip C6 processor
sets the flags the same as does the Pentium processor.
In some cases where the various Pentium-compatible processors
differ on these undefined flag settings, the IDT WinChip C6
processor settings differ from the Pentium processor settings.

CMPXCHG8B
The CMPXCHG8B instruction is supported by the IDT WinChip
C6 processor. However, it is not reported as being supported in
the CPUID instruction. This was required to maintain
compatibility with some Windows NT versions.

B.4 FLOATING-POINT COMPATIBILITY

Transcendental Accuracy
The results of the transcendental instructions (FSIN, etc.) are
slightly different among the various x86 processors—including
between the Intel486 processor and the Pentium processor. The
IDT WinChip C6 processor typically gets results within one-half
ULP (units in the lower position) average and within one ULP
worst-case of the correct mathematical result within the reduced
argument range.

Undefined Condition Bits

The Intel Pentium documentation defines the setting of the FPU
Status Word condition bits for several instructions as
“undefined”. In some cases the setting of undefined flags is
consistent among Pentium-compatible processors and some
programs depend upon these settings. In these cases, the IDT
WinChip C6 processor sets the flags the same as does the
Pentium processor.

COMPATIBILITY B-5
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

In some cases where the various Pentium-compatible processors

differ on these undefined flag settings, the IDT WinChip C6
processor settings differ from the Pentium processor settings.

B-6 COMPATIBILITY
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

APPENDIX C. PROCESSOR ERRATA

C.1 INTRODUCTION
Appendix B describes the expected differences between the IDT
WinChip C6 processor and the Intel Pentium processor. This
Appendix A describes the IDT WinChip C6 processor errata:
differences between the actual IDT WinChip C6 processor
behavior and the expected results.
There are two current versions or steppings of the IDT WinChip
C6 processor which have different errata: steppings 0 and 1.
These can be identified by the stepping code returned by the
CPUID instruction or in EDX following Reset.
Tables C-1 and C-2 describe the codes used to define the status
and action plan of each individual erratum.

Table C-1
Status Description
Code
X This version of the processor has the erratum
N/A This erratum does not apply to this version of the processor
Fixed in this version of processor

Table C-2
Plan Description
Code
Fixed This erratum is fixed in the latest version of the processor
Fix This erratum may be fixed in some future version of the
processor
NoFix There are no plans to fix this erratum in future steppings of
the IDT WinChip C6 processor

ERRATA C-1
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

The IDT WinChip C6 processor errata described here is

complete and is described in considerable detail. This detailed
exposure is consistent with the Centaur philosophy of openness.
However, the errata do not, as far as we know, represent a real
compatibility exposure. Microsoft and XXCAL did not encounter
these during compatibility testing. Two of these errata were
found during our extensive beta/system test. The rest were
found only by specialized design-verification tests, which
perform unnatural acts.

Table C-3 summarizes the errata and provides the codes which
describes the status and action plan for each individual
erratum. The details of each erratum are described in
subsequent sections.
Table C-3
ID Stepping Plans Errata
0 1

F-1 X X NoFix FPU environment may be different from P55 in

certain cases
F-2 X X NoFix FPU transcendental may return a denormal value
without underflow
F-3 X Fixed FPU FSINCOS or FCOS may cause DE to be
incorrectly set
F-4 X Fixed FPU may slow down integer performance with
incorrect pipeline slips
F-5 X X NoFix FPU FCOM,FCOMP of denormal memory operand
may not set flags correctly
F-6 X Fixed FPU FPREM of pseudodenormal does not convert
result to Normal
F-7 X Fixed FPU instruction may result in spurious exception
under certain conditions
F-8 X X Fix FPU FRNDINT does not round 2^63 – 0.5 correctly
F-9 X X Fix FPU FPREM1 may get wrong sign
F-10 X X Fix FPU FIST,FISTP rounds certain numbers
incorrectly
M-1 X X NoFix MMX instruction followed by read of FPSW may
get wrong top of stack
I-1 X X NoFix INTR is missed if deasserted while processor
executes consecutive STIs
I-2 X X NoFix Certain invalid instructions may result in page
faults instead of INT6

C-2 ERRATA
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

ID Stepping Plans Errata

0 1

I-3 X X NoFix BUSCHK# does not work as in P54 in certain

cases
I-4 X Fixed Wrong EIP pushed when INTR or NMI wakes
processor after unmasked FPU exception
I-5 X Fixed INVLPG with linear address close to 4GB may
incorrectly GP fault
I-6 X Fixed Self-modifying code may not be detected as in P54
I-7 X Fixed Time Stamp Counter stops counting when
processor in low power standby
I-8 X Fixed TR12.AHD does not disable Auto HALT
Powerdown as it should
I-9 X X NoFix Inconsistent fault when length limit exceeded with
invalid LOCK prefix
I-10 X X Fix BIST return code (EAX) may indicate false failures
I-11 X X Fix EFLAGS.TF incorrectly causes trap on entry to
SMM
I-12 X X Fix RDTSC may return incoherent values in EDX:EAX
I-13 X X Fix RDTSC hangs when certain event counters are
used
I-14 X X Fix EFLAGS.RF incorrectly cleared after external
interrupt
I-15 X X Fix SMI# during REP string does not set EFLAGS.RF
I-16 X X Fix Call gate may incorrectly cause page protection
fault
B-1 X Fixed MCR_CTRL[4] will not enable weak write ordering
as it should
B-2 X Fixed Data bits with inactive byte enables may differ
from P54 during I/O writes
B-3 X Fixed TR12.CI does not inhibit cache line fills as it should
B-4 X X Fix Instruction cache does not detect internal parity
error
B-5 X X Fix HIT# may incorrectly stay asserted if snoop occurs
during INVD / WBINVD
B-6 X X Fix Memory read with disabled cache may result in
cache line writeback and/or invalidate

ERRATA C-3
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

C.2 FLOATING-POINT ERRATA

F-1. FPU environment may be different from P55 in

certain cases
PROBLEM: An FSTP QWORD that crosses a page
boundary and gets a page fault will set the floating point
environment, technically it should not. An FRSTOR that
crosses a segment limit of 4GB will partially load floating-
point state.
IMPLICATION: These errata are not believed to occur in
any application or operating system.
WORKAROUND: None

F-2. FPU transcendental may return a denormal value

without underflow
PROBLEM: The transcendental instructions (FSIN,
FCOS, FSINCOS, FPTAN, FPATAN, FYL2X, FYL2XP1,
F2XM1) may, in some cases, return a denormal value
without underflow. This can occur whenever the internal
approximation algorithm computes a precise denormal
result (i.e. no loss of precision occurred in the computation
of the approximation).
In these cases, the transcendental instructions behave like
the arithmetic instructions. This is technically incorrect,
since transcendental approximations are inherently
inexact.
If the C6 transcendental approximation algorithm
computes an exact denormal result, UE is not set in FPSW,
and if UE is unmasked, the unmasked underflow response
does not occur.
Examples:
FSIN(denormal x) = x with no underflow
FYL2X(2, denormal y) = y with no underflow

IMPLICATION: Applications that run with the underflow

exception masked, and do not check the UE bit in FPSW
(virtually all applications fall into this category) are
unaffected.

C-4 ERRATA
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

WORKAROUND: In applications that do respond to

underflow, this condition can be detected by examining
result of a transcendental operation with FXAM to see if it
is denormal, and if it is, handling that case as underflow.
For applications that expect the unmasked underflow
response, the translation of the value to the middle of the
normal range can computed without loss of precision in
user code:
if (denormal result)
result = result * 2 ^ 24576

F-3. FPU FSINCOS or FCOS may cause DE to be

incorrectly set
PROBLEM: FSINCOS or FCOS on some very small
normal values can cause DE to be incorrectly set, which
could cause a false exception if DE is unmasked.
IMPLICATION: Since few programs run with DE
unmasked, especially among those that use transcendental
instructions, this situation is extremely unlikely to cause
problems.
WORKAROUND: Run with DE exception masked.

F-4. FPU may slow down integer performance with

incorrect pipeline slips
PROBLEM: If Dynamic Power Management is enabled
(the default condition) a floating-point status bit may be
incorrectly captured when clocks are being deasserted to
the FPU. There are other conditions internal to the
processor pipeline required for this erratum, which makes
it an unlikely event.
IMPLICATION: The FPU will falsely stall the processor
resulting in one clock added per micro-instruction until
another floating-point instruction, integer divide, or
integer multiply is executed.
WORKAROUND: Disable Dynamic Power Management.

ERRATA C-5
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

F-5. FPU FCOM,FCOMP of denormal memory operand

may not set flags correctly
PROBLEM: The C0, C2, C3 flags in the floating-point
status word may not be set when an FCOM instruction
results in a stack fault. This erratum will occur when the
following conditions are met:
• One of the FCOM operands is in memory (single- or
double-real memory operand)
• the memory operand is Denormal
• the register operand causes a stack fault (underflow)
• the invalid operation exception is unmasked
The correct processor response is to set C0, C2, C3 flags
indicating that the operands were “not comparable.”
IMPLICATION: It is unlikely that the exception handler,
when dealing with an invalid operation exception, would
need the C-bits set as a result of the faulting FCOM. The
exception handler will update the floating-point stack and
restart the faulting FCOM or terminate the faulting
process. It is therefore unlikely that this erratum will
cause compatibility problems.
WORKAROUND: Run with invalid operation exception
masked.

C-6 ERRATA
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

F-6. FPU FPREM of pseudodenormal does not convert

result to Normal
PROBLEM: Pseudodenormals are numbers in the
extended format with a zero in the exponent but a ‘1’ in the
most significant bit of the significand (in the extended
format the leading bit is explicitly stored). The numeric
value corresponding to denormals/pseudodenormals is
computed as for normals except the exponent is assumed to
be 1 (instead of the encoded zero). This means that, by
definition, the numeric value of pseudodenormals overlaps
the range of Normals for which exponent is 1. Numbers in
this range, when represented in the extended format, could
be either Normals (exponent of 1) or pseudodenormals
(zero exponent). The x86 architecture allows
pseudodenormals as operands but requires that the
processor generate a Normal rather than a
pseudodenormal. Due to this erratum the FPREM
instruction will return a pseudodenormal when the input is
a pseudodenormal. The correct processor response is to
convert the operand into a Normal (make the exponent 1).
IMPLICATION: Pseudodenormals are rarely encountered
in applications, and it is unlikely that they would be the
input to FPREM. The processor response in this case,
although architecturally incorrect, is numerically correct.
If the pseudodenormal result of FPREM were to be used in
a subsequent computation the answer would be correct. It
is therefore not expected to cause problems in any
application.
WORKAROUND: None.

F-7. FPU instruction may result in spurious exception

under certain conditions
PROBLEM: The processor may generate a spurious
floating-point exception under the following conditions:
• An instruction which will fault or modify the
instruction stream begins execution
• An FIST or FISTP instruction follows the faulting
instruction

ERRATA C-7
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

• The FIST or FISTP instruction determines there is an

overflow
• The fault (or self-modifying store) from the preceding
instruction flushes the pipeline
• A subsequent FPU instruction will react to the
lingering overflow condition detected by the flushed
FIST, possibly generating a spurious floating-point
exception
IMPLICATION: Applications which use FIST or FISTP
with data which will overflow and which are preceded by
instructions which fault or store into the instruction
stream are at risk of getting spurious floating-point
exceptions.
WORKAROUND: This erratum can be avoided by placing
the FIST or FISTP instruction in a subroutine, forcing the
processor to branch prior to executing it and thereby
clearing the pipeline.

F-8. FPU FRNDINT does not round 2^63 – 0.5 correctly

PROBLEM: The FRNDINT instruction will not produce
the correct result when the following conditions are met:
• the operand of FRNDINT is 2^63 – 0.5 and
• the rounding control in the Floating Point control word
is either ‘up’ or ‘to nearest’
The processor will return 2^63 – 1 and will not set either
FPSW.PE or FPSW.C1. The correct processor response is
to store 2^63 and set both FPSW.PE and FPSW.C1
indicating that the result is imprecise and that the
processor rounded up.
IMPLICATION: Applications relying on the FRNDINT
instruction to round this special value will fail. This
erratum has not been determined to cause any problems
during compatibility testing.
WORKAROUND: Either of these workarounds will avoid
the problem:
• Set the rounding control in the Floating Point control
word (FPCW.RC) to round ‘toward zero’ or ‘down’.

C-8 ERRATA
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

• Check the operand and avoid executing FRNDINT

when it is 2^63 – 0.5.

F-9. FPU FPREM1 may get wrong sign bit

PROBLEM: In certain circumstances, the FPREM1
instruction can incorrectly round the intermediate
quotient, and therefore return a result with an incorrect
sign. This occurs in the following conditions:
• a preceding FPU operation has set FPSW.PE = 1 and
• the quotient ST0/ST1 can be represented precisely and
• the fractional part of the quotient ST0/ST1 is
EXACTLY 0.5 and
• the integer part of the quotient is odd.
IMPLICATION: For example, with FPSW.PE = 1, ST0 =
3.0, ST1 = 2.0 FPREM1 returns -1.0 instead of 1.0 (for ST0
= 5.0, ST1 = 2.0 the result is correct because the integer
part of the quotient is even).
The instructions FPREM, FSIN, FCOS, FPTAN (which use
similar reduction algorithms) are unaffected.
WORKAROUND: The situations in which the error occurs
are rare, and an incorrectly signed result is unlikely to
cause problems numerically in an algorithm employing
FPREM1 for reduction. However, if IEEE-compliant
rounding at interval endpoints is desired, FPREM1 will
behave correctly if FPSW.PE is clear when the FPREM1
instruction is executed. For example, the code sequence:
FCLEX
FPREM1
will ensure IEEE-compliant behavior.

ERRATA C-9
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

F-10. FPU FIST,FISTP do not handle overflow correctly

PROBLEM: The FIST and FISTP instructions will not
handle overflow correctly when the following conditions are
met:
CASE 1:
• the operand of FIST/FISTP is positive and
• the rounding control in the Floating Point control word
is either ‘up’ or ‘to nearest’ and
• the rounded operand does not fit in the destination
integer and
• the rounded operand would have fit in the destination
integer if rounding had been in the other direction
The processor sets FPSW.OE to indicate overflow, and
raises Overflow Exception if it is unmasked. The correct
processor response is to set FPSW.IE to indicate Invalid
Operation exception, and raise Invalid Operation exception
if it is unmasked. The following table specifies the range of
operands ‘x’ for which the processor indicates the wrong
exception:
FPCW.RC=’up’ FPCW.RC=’to nearest’
FIST[P] 2^15 – 1 < x < 2^15 2^15 – 0.5 <= x < 2^15
m16int
FIST[P] 2^31 – 1 < x < 2^31 2^31 – 0.5 <= x < 2^31
m32int
FISTP x = 2^63 – 0.5 x = 2^63 – 0.5
m64int

CASE 2:
• the operand of FIST/FISTP is negative and
• the rounding control in the Floating Point control word
is either ‘up’ or ‘to nearest’ or ‘toward zero’ and
• the operand is less than the maximum negative value
that fits in the destination integer

C-10 ERRATA
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

• the rounded operand is the maximum negative value

that fits in the destination integer
The processor sets FPSW.IE, and raises Invalid Operation
exception if it is unmasked (in addition, if the instruction is
FISTP, the FPTW may incorrectly be cleared). The correct
processor response is to set FPSW.PE to indicate Precision
exception, and raise Precision exception if it is unmasked.
The following table specifies the range of operands ‘x’ for
which the processor indicates the wrong exception:
FPCW.RC=’up’ FPCW.RC=’to nearest’
FIST[P] -2^15 – 1 < x < -2^15 -2^15 – 0.5 <= x < -2^15
m16int
FIST[P] -2^31 – 1 < x < -2^31 -2^31 – 0.5 <= x < -2^31
m32int
IMPLICATION: Applications relying on the proper
exception signaled in the floating-point status word when
FIST/FISTP overflows may fail. This erratum has not been
determined to cause any problems during compatibility
testing.
WORKAROUND: Any of these workarounds will avoid
the problem:
• Set the rounding control in the Floating Point control
word (FPCW.RC) to round ‘down’.
• Check the range of the operand and avoid executing
FIST/FISTP when the input operand is greater than
MAXINT-1 or less than MAXNEG.

C.3 MMX ERRATA

M-1. MMX instruction followed by read of FPSW may get
wrong top of stack
PROBLEM: An MMX instruction followed immediately
(without an EMMS) by a read of the Floating Point Status
Word (e.g. FSTSW,FDECSTP,FINCSTP,FSAVE) may
report the incorrect top of stack (TOS). The correct
response is for MMX instructions to clear the TOS.

ERRATA C-11
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

IMPLICATION: MMX applications should not encounter

this erratum. Under the prescribed method of executing
MMX instructions after Floating Point routines, Section
3.3.2 of the Intel MMX specification states that EMMS
must be used as a barrier between MMX/FPU instructions,
or else errors may result.
WORKAROUND: Use EMMS between MMX and FPU
instructions as prescribed by the Intel MMX specification.

C-12 ERRATA
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

C.4 INTEGER ERRATA

I-1. INTR is missed if deasserted while processor

executes consecutive STIs
PROBLEM: If the processor is executing a sequence of
more than one STI and an INTR is asserted and then
deasserted before the processor ends the sequence of STIs
then INTR will not be recognized.
IMPLICATION:
This situation should not happen on a PC system because:
• INTR is level sensitive, the interrupt controller must
latch interrupts from devices and assert INTR until the
processor acknowledges.
• There is no reason for software to use consecutive STIs.
WORKAROUND: None needed, see above.

I-2. Certain invalid instructions may result in page faults

instead of INT6
PROBLEM: The illegal opcode 0F, 71h 72h and 73h with
ModR/m memory (Group A MMX shift) is not flagged as an
invalid opcode if the instruction crosses a page boundary
and the second page faults, the IDT-C6 processor presents
a page fault exception instead
IMPLICATION: If this were to happen the operating
system will allocate the second page and resume the
faulting task which would then get the invalid opcode
exception (INT6).
WORKAROUND: None needed, see above.

I-3. BUSCHK# does not work as in P54 in certain cases

PROBLEM: If BUSHCK# and SMI# are asserted while in
HALT state the processor correctly takes BUSCHK# first
but Auto Halt Restart slot in SMRAM is not set when
SMM is entered. A similar BUSCHK# erratum is that the
ESI I/O restart slot may be incorrect if BUSCHK# and
SMI# are asserted during an I/O operation.

ERRATA C-13
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

IMPLICATION: BUSCHK# is not used in PC systems.

WORKAROUND: None needed in a PC system, see above.

I-4. Wrong EIP pushed when INTR or NMI wakes

processor after unmasked FPU exception
PROBLEM: The processor will push the wrong EIP on the
stack when it services an external interrupt under the
following conditions:
• CR0.NE is 0, indicating the processor should not raise
INT16 on unmasked FPU exceptions
• an unmasked FPU exception occurs
• IGNNE# is inactive
• the processor does not clear or mask the FPU exception
before the next FPU instruction which causes the
processor to wait for an external interrupt
• at this point the processor waits for an external
interrupt
• INTR or NMI is asserted
• Due to this erratum, the processor pushes the wrong
EIP. The correct response is to push the EIP of the FPU
instruction which caused the processor to wait for an
external interrupt
IMPLICATION: DOS applications that have unmasked
FPU exceptions and provide their own interrupt handler
are subject to this erratum.
WORKAROUND: None

I-5. INVLPG with linear address close to 4GB may

incorrectly GP fault
PROBLEM: A GP fault will result when INVLPG specifies
a linear address within the last few bytes of the 32-bit
address limit. The specific linear address subject to this
erratum depends on the operand size:
• For 16-bit operand size the faulting linear address is
0xFFFFFFFF

C-14 ERRATA
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

• For 32-bit operand size the faulting linear address is in

the range {0xFFFFFFFD, 0xFFFFFFFE,0xFFFFFFFF}.
This erratum is independent of the address size specified
by the INVLPG.
IMPLICATION: This erratum has not been observed in
any operating system.
WORKAROUND: An operating system can avoid this
erratum by ensuring that the linear address specified by
INVLPG is not in the faulting range. Masking the address
so it points to the first byte of the page would meet this
requirement.

I-6. Self-modifying code may not be detected as in P54

PROBLEM: The processor may not detect self-modifying
code under the following conditions:
• An instruction stores to memory either 1 or 2 bytes (no
more)
• The destination byte(s) are contained within the last 3
bytes of a cache line
• The instruction storing to memory is within 3 cache
lines of the target
• The target sequentially follows the instruction storing
to memory with no branches separating them
• Due to this erratum, the processor may have pre-
fetched the target and not detect that it was modified,
causing it to execute the previous contents at the target.
The correct response is for the instruction stream to reflect
all preceding stores.
IMPLICATION: This erratum does not manifest itself in
any known applications.
WORKAROUND: Issue a branch after modifying the
target instruction.

ERRATA C-15
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

I-7. Time Stamp Counter stops counting when

processor in low power standby
PROBLEM: The processor will incorrectly stop
incrementing the Time Stamp Counter when it enters the
low power standby state, i.e. under the following
conditions:
• STOP GRANT (STPCLK# asserted)
• SHUTDOWN
• the HALT instruction (Auto HALT Powerdown)
• There is an unmasked floating-point error, CR0.NE is
'0', IGNNE# is deasserted
The correct response is for the Time Stamp Counter to
continue incrementing at all times.
IMPLICATION: This erratum manifests itself as a
spontaneous "Divide by zero" message when Linux is idle
for some time.
WORKAROUND: None.

I-8. TR12.AHD does not disable Auto HALT Powerdown

as it should
PROBLEM: When TR12.AHD (Auto Halt Disable) is set to
'1' the processor should not Powerdown, but due to this
erratum it does.
IMPLICATION: This erratum has not been determined to
cause any problems during compatibility testing.
WORKAROUND: None.

I-9. Inconsistent fault when length limit exceeded with

invalid LOCK prefix
PROBLEM: The LOCK prefix (0xF0) is valid only for
certain instructions. The LOCK prefix is not valid, for
example, when the operands are all in registers. The
correct processor response for such an invalid use of the
LOCK prefix is to generate an Invalid Opcode exception
(INT 6).

C-16 ERRATA
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

The use of redundant prefixes could result in the

instruction length exceeding 15 bytes; the correct processor
response is to generate a General Protection fault (INT 13).
When an instruction has an invalid LOCK prefix and the
length exceeds 15 bytes the resulting exception (INT 6 or
INT 13) is implementation-dependent. The Pentium
processor (P54) responds with INT 6 whereas the Pentium
Processor with MMX Technology (P55) responds with INT
13. The IDT WinChip C6 processor response is not
consistent, for some opcodes it responds with INT 6 (like a
P54) and for others with INT 13 (like a P55).
IMPLICATION: This erratum has not been determined to
cause any problems during compatibility testing.
WORKAROUND: Do not rely on implementation-
dependent behavior.

I-10. BIST return code (EAX) may indicate false failures

PROBLEM: The Built-in Self Test (BIST) can be
requested as part of the IDT WinChip C6 processor reset
sequence (INIT asserted as RESET deasserted). The BIST
result is indicated by a code in EAX. Normally EAX is zero
after reset. A ‘zero’ in EAX after BIST Reset means that no
failures were detected. Any value other than ‘zero’
indicates an error has occurred during BIST. Due to this
erratum the processor may return a value other than ‘zero’
in EAX, even though no failures were detected. There are
two errors in BIST which compound this problem:
• EAX is not set to zero by the processor prior to BIST,
therefore random bits may be set
• Bit 3 of EAX may be set to ‘one’, falsely indicating an
error that does not exist
IMPLICATION: PC systems typically don’t invoke BIST,
therefore this erratum has not been determined to cause
any problems during compatibility testing.
WORKAROUND: In systems where BIST is desired the
following sequence will avoid the problem:
• RESET the processor normally (this sets EAX to ‘zero’).
• Indicate in volatile RAM that BIST will be requested

ERRATA C-17
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

• Request BIST
• In the RESET code check the volatile RAM to see if
BIST was requested (should be true) and if so, mask bit
3 of EAX and compare with ‘zero’.

I-11. EFLAGS.TF incorrectly causes trap on entry to

SMM
PROBLEM: The EFLAGS.TF bit may be set by a POPF,
IRET, RSM or task switch. It is used to single-step
instructions. The processor raises a trap exception upon
completing the next instruction and invokes the INT1
debug handler. If SMI# is asserted and recognized after
EFLAGS.TF is set but before the instruction that should be
single-stepped is executed the processor should enter SMM
and save EFLAGS (with TF set) in SMRAM and then clear
EFLAGS.TF and begin the SMM handler code without
raising INT1 trapped exception. Due to this erratum the
processor raises INT1 after making the transition to SMM.
IMPLICATION: SMI# is a relatively infrequent interrupt,
and EFLAGS.TF is typically used by debuggers, so it is
also rare. However, this erratum has been observed during
compatibility testing. The game “Wings of Glory” in a
system with a USB keyboard fails due to this erratum.
WORKAROUND: None identified at this time.

I-12. RDTSC may return incoherent values in EDX:EAX

PROBLEM: The Time Stamp Counter (TSC) is a 64 bit
register that counts processor clocks. Programs read the
TSC using the RDTSC instruction or RDMSR with
ECX=10h. The upper 32 bits are returned in EDX and the
lower 32 bits are returned in EAX. In the IDT WinChip C6
the TSC is implemented by using the lower 32 bits of both
of the performance counters, CTR0 and CTR1. Due to this
erratum it is possible that reading the TSC returns a value
in EDX that is inconsistent with the value in EAX. This
happens when the lower 32 bits of the TSC (implemented
in CTR0) are close to 0xFFFFFFFF. If is possible for the
processor to load EAX with the lower 32 bits just after they
roll over to 0x00000000 and then load EDX with the upper
32 bits before they show the carry out of the lower 32 bits.

C-18 ERRATA
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

IMPLICATION: System software might fail if it expects

the 64 bit TSC to increase monotonically and it executes
RDTSC when the lower 32 bits of the TSC are all ones.
The probability that the RDTSC instruction execution will
coincide with the lower 32 bits of the TSC being all ones is
very low (1 in 4 billion).
WORKAROUND: This erratum can be detected by
comparing EAX to ‘zero’ after reading the TSC. The
erratum can be avoided by reading the TSC again if EAX is
‘zero’.

I-13. RDTSC hangs when certain event counters are

used
PROBLEM: The Time Stamp Counter (TSC) is a 64 bit
register that counts processor clocks. Programs read the
TSC using the RDTSC instruction or RDMSR with
ECX=0x10. The upper 32 bits are returned in EDX and
the lower 32 bits are returned in EAX. In the IDT
WinChip C6 the TSC is implemented by using the lower 32
bits of both of the performance counters, CTR0 and CTR1.
Due to this erratum it is possible that reading the TSC will
cause the processor to hang. This happens when the
CESR.CTR1CTL has been changed by system software
from its default (0x00) to any of the following values (all of
which are undefined):
• 0x10
• 0x15
• 0x21
IMPLICATION: System software might cause the
processor to hang if undefined event counter control values
are loaded into the CESR.CTR1CTL field.
WORKAROUND: System software should avoid
undefined event counter control values.

ERRATA C-19
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

I-14. EFLAGS.RF incorrectly cleared after external

interrupt
PROBLEM: EFLAGS.RF is used to signal that the
processor should ignore an instruction breakpoint at the
target of the current control transfer. It could be set by one
of the following:
• An IRETD when it pops EFLAGS from the stack
• An RSM when it reads EFLAGS from SMRAM
• A task switch when it reads EFLAGS from the incoming
32 bit TSS
If at the end of such a control transfer that sets
EFLAGS.RF there is a pending external interrupt then the
processor incorrectly clears EFLAGS.RF. Depending on
the interrupt and how it is handled this incorrect value of
EFLAGS is then:
• pushed on the stack
• saved in SMRAM
• stored in the outgoing 32 bit TSS
IMPLICATION: This erratum manifests itself as an
instruction breakpoint that gets taken twice. This erratum
is unlikely to occur, provided the debugger does not disable
external interrupts for a prolonged period of time
immediately prior to resuming the interrupted program.
WORKAROUND: When this erratum occurs the user may
or may not notice that the program has made no forward
progress. The state of the program under debug is not
altered in any way, so resuming execution is all that is
necessary.

C-20 ERRATA
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

I-15. SMI# during REP string does not set EFLAGS.RF

PROBLEM: EFLAGS.RF is used to signal that the
processor should ignore an instruction breakpoint at the
target of the current control transfer. When a REP string
instruction is interrupted the processor should save an
image of EFLAGS that has EFLAGS.RF set indicating that
if there is an instruction breakpoint at this instruction, it
as already been taken. This is necessary because the EIP
saved points to the REP string instruction so that it can
resume after the interrupt is serviced. Due to this erratum
the processor does not set EFLAGS.RF in SMRAM when
SMI# is asserted while a REP string is executing.
IMPLICATION: This erratum manifests itself as an
instruction breakpoint that gets taken twice. This erratum
is unlikely to occur because SMI# is not frequent and it is
unlikely that there is an instruction breakpoint set at a
REP string that happens to be interrupted by SMI#.
WORKAROUND: When this erratum occurs the user may
or may not notice that an instruction breakpoint is taken
again before the REP string reaches the end. The state of
the program under debug is not altered in any way, so
resuming execution is all that is necessary.

I-16. Call gate may incorrectly cause page protection

fault
PROBLEM: A call gate may be used to transfer control to
a procedure in a segment at a higher privilege level
(referred to as an inner level, which has numerically lower
DPL in the code segment descriptor). Privilege levels 0
(highest privilege) through 2 are considered “supervisor”
for purposes of page protection. Privilege level 3 (least
privileged) is considered “user” privilege. The call gate
descriptor specifies how many parameters to copy from the
caller’s (outer) stack to the stack at the higher privilege
level (inner stack). Due to this erratum the processor will
incorrectly raise a page protection fault when all the
following conditions are met:
• The processor is in protected mode (CR0.PE=1) and
• paging is enabled (CR0.PG=1) and

ERRATA C-21
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

• the current privilege level (CPL) is 1 or 2 and

• the processor transfers control through a call gate to an
inner level (from 2 to 1 or from 2 to 0 or from 1 to 0) and
• there is at least 1 parameter specified in the call gate
descriptor and
• the ESP of the caller’s (outer) stack is misaligned such
that the offset of the least significant byte of the first
parameter modulo 4 is 3 (ESP after first parameter
pushed modulo 4 = 3) and
• at least one of the bytes of the parameter(s) on the outer
stack is in a supervisor-level page
Due to this erratum the processor will raise a page
protection fault while attempting to copy the parameter(s)
from the outer stack to the inner stack.
IMPLICATION: This erratum is unlikely to occur given
that operating systems properly align the stack pointer (to
avoid performance degradation) and given that call gates
with parameters are rare. This erratum has not been
observed in any operating system during compatibility
testing.
WORKAROUND: This erratum cannot be encountered by
a user-level procedure (executing at privilege level 3). To
ensure that this erratum is not encountered in supervisor-
level procedures it is sufficient to use a properly aligned
stack.

C.5 BUS ERRATA

B-1. MCR_CTRL[4] will not enable weak write ordering
as it should
PROBLEM: Weak write ordering is a performance
feature of the IDT-C6 processor that is controlled by bit 4
of the MCR_CTRL machine specific register in conjunction
with the Memory Range Registers. Due to this erratum
the processor may end up with a cache line in the exclusive
state instead of modified, leading to stale system memory
contents.

C-22 ERRATA
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

IMPLICATION: Memory Range Registers should not

specify weak write ordering, and therefore performance of
the IDT-C6 processor may not be optimal.
WORKAROUND: Get patch from Centaur to allow
Memory Range Registers to specify weak write ordering.

B-2. Data bits with inactive byte enables may differ from
P55 during I/O writes
PROBLEM: Bits on the data bus with inactive byte
enables have undefined values. During an I/O write bus
cycle for an 8-bit or 16-bit operand P54 drives '0' for data
bits that are not part of the operand (the operand is zero-
extended and rotated so that it is properly aligned on the
bus).
IMPLICATION: This erratum is known to cause problems
in a system with an Opti Firestar chip set . In this case the
ESDI device driver has a long delay because the secondary
PCI controller indicates a busy status indefinitely.
WORKAROUND: For P55 compatible behavior zero-
extend the operand into EAX and use register operands for
the OUT instruction (use OUT DX,AX or OUT DX,AL
instead of OUTS).
Example:
MOVZX EAX,BX ; Zero-extend to 32-bits
OUT DX,AX ; Do not use OUTS

B-3. TR12.CI does not inhibit cache line fills as it should

PROBLEM: When TR12.CI (Cache Inhibit) is set to '1' all
cache line fills should be inhibited; during cache misses the
processor should access memory using single transfer
cycles. Due to this erratum TR12.CI is ignored and the
caches operate normally regardless of the setting of
TR12.CI.

ERRATA C-23
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

IMPLICATION: If software (like BIOS) relies on TR12.CI

to test the second level cache, and if the second level cache
is defective, then the defect may go unnoticed. If TR12.CI
is used to increase performance when accessing code or
data (that will not be referenced again soon), then
performance might be adversely affected if a cache line
that is frequently referenced is replaced by a cache line
that is less frequently referenced. This erratum has not
caused problems in compatibility testing.

WORKAROUND: Second level cache tests could be

written to avoid TR12.CI, although this might require
flushing the internal caches several times and would
therefore increase the time required to perform the second
level cache test.

B-4. Instruction cache does not detect internal parity

errors
PROBLEM: The instruction cache should compute and
maintain a parity bit for each 32 bits. These parity bits
should be compared to the computed parity whenever the
instruction cache is accessed. If a stored parity bit does not
match the computed parity bit the correct processor
response is to assert IERR# and handle the exception as
specified by machine–specific registers (the default is
shutdown). Due to this erratum internal parity errors in
the instruction cache are not detected.
IMPLICATION: In the unlikely event that a bit stored in
the instruction cache spontaneously changes value it may
not be detected; the result is unpredictable.

WORKAROUND: None.

B-5. HIT# may incorrectly stay asserted if snoop occurs

during INVD / WBINVD
PROBLEM: The HIT# signal is used to indicate that an
inquire cycle (external snoop) hits a matching address
within a processor cache. The HIT# signal transitions only
as a response to an inquire cycle. Due to this erratum the
HIT# signal remains asserted when it should have been
deasserted as a result of this sequence of events:

C-24 ERRATA
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

• The HIT# signal is asserted as a response to an inquire

cycle
• An INVD or WBINVD instruction begins execution
• An inquire cycle occurs during the INVD/WBINVD
instruction
• At this point the HIT# signal should transition to
deasserted, but due to this erratum it remains asserted
IMPLICATION: Most chipsets in PC systems do not
respond to the HIT# signal, and most operating systems do
not use the INVD or WBINVD instruction. This erratum
has not been observed during compatibility testing.

WORKAROUND: None needed in a PC system, see above.

B-6. Memory read with disabled cache may result in

cache line writeback and/or invalidate
PROBLEM: The processor caches are disabled when
either CR0.CD=1 or TR12.CI=1. When caches are disabled
the processor should not perform cache line fill operations;
that is, no new cache lines should be allocated. If an access
misses in the cache when caches are disabled, the correct
processor response is to issue a single cycle transfer (two
for a data access that crosses an 8-byte boundary).
Typically no cache line writeback or invalidates occur when
caches are disabled. There are special cases where
writebacks and/or invalidates occur even though caches are
disabled.
Coherency requirements dictate that the processor
writeback and/or invalidate a cache line when caches are
disabled under the following circumstances (compatible
with Pentium):
• CR0.NW=0 and an external snoop hits in the cache, the
cache line is written back if modified and invalidated.
• A write misses in the data cache but hits in the
instruction cache; the instruction cache line is
invalidated.

ERRATA C-25
 Preliminary Information
March 1998 IDT WINCHIP TM C6TM PROCESSOR DATA BOOK

• An instruction fetch misses in the instruction cache but

hits a modified line in the data cache; the cache line is
written back and invalidated.
• A locked cycle (read-modify-write sequence) hits a line
in the data cache; the cache line is written back if
modified and invalidated.
The cache organization of the IDT WinChip C6 processor
may result in a writeback and/or invalidate when caches
are disabled, a behavior different from Pentium processor,
but not considered errata:
• A read or write that hits in the cache uses a virtual
alias. This happens when the physical address hits in
the cache but there are different values in either bit 13
or 12 of the linear address. If the aliased line is
modified then it is written back. The line is
invalidated.
There is one remaining case where the processor with
disabled caches will writeback and/or invalidate cache
lines as a response to a memory read. The behavior in this
case is not compatible with Pentium and is the reason for
this erratum. The improper writeback and/or invalidate
will occur when:
• A read that misses in the cache occurs to an index
whose LRU points to a valid line. If the line is modified
then it is written back. The line is invalidated. This is
a result of the processor ensuring it has an available
cache line to allocate and store the data from the bus
even though caches are disabled.
IMPLICATION: This erratum has not been determined to
cause any problems during compatibility testing. This
erratum would have harmful effects only if the processor
modifies cache lines for which there is no corresponding
writable memory enabled in the system at the time, such
that the data from a writeback operation could not be
retrieved with a later read cycle.
WORKAROUND: When operating with caches disabled,
ensure that the system has writable memory at the
physical address corresponding to all modified cache lines.

C-26 ERRATA