See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/201987790

A New Trend for CISC and RISC Architectures

Article in Asian Journal of Information Technology · January 2007

CITATION READS
1 16,977

2 authors:

Aws Fida El-Din Hasan Krad

Qatar University Qatar University
14 PUBLICATIONS 9 CITATIONS 20 PUBLICATIONS 45 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Reducing HTTP request times through utilizing BGP. View project

All content following this page was uploaded by Aws Fida El-Din on 20 April 2014.

The user has requested enhancement of the downloaded file.

 A New Trend for CISC and RISC Architectures

Hasan Krad and Aws Yousif Al-Taie

Department of Computer Science & Engineering

College of Engineering
Qatar University
{hkrad, altaie}@qu.edu.qa

ABSTRACT
The comparative study between CISC (Complex Instruction Set Computer) and RISC
(Reduce Instruction Set Computers) has been a well known research area for many
years. In our paper, we try to address the new trend of these two architectures, which
is CRISC (Complex-Reduce Instruction Set Computer). We chose the Intel Core Duo
processor, Intel's most recent processor, to be the focus of our study. The Core Duo
processor features will be highlighted, focusing on pipelining stages, clock speed,
number of transistors, instruction set architecture (ISA), and the improvement in
cache technology.

1. Introduction
From the architecture point of view, the microprocessor chips can be classified into
two categories: Complex Instruction Set Computers (CISC) and Reduce Instruction
Set Computers (RISC). In either case, the objective is to improve system
performance. The debates between these two architectures made this research area
very interesting, challenging, and some times confusing.

CISC (Complex Instruction Set Computer) computers are based on a complex

instruction set in which instructions are executed by microcode. Microcode allows
developers to change hardware designs and still maintain backward compatibility with
instructions for earlier computers by changing only the microcode, thus make a
complex instruction set possible and flexible. Although CISC designs allow a lot of
hardware flexibility, the supporting of microcode slows microprocessor performance
because of the number of operations that must be performed to execute each CISC
instruction. A CISC instruction set typically includes many instructions with different
sizes and execution cycles, which makes CISC instructions harder to pipeline [1].

From the 60's CISC microprocessors became prevalent, each successive processor
having more and more complicated hardware and more and more complex instruction
sets. This trend started from Intel 80486, Pentium MMX to Pentium III.

RISC (Reduced Instruction Set Computer) chips evolved around the mid-1970 as a
reaction at CISC chips. In 70's, John Cocke at IBM's T.J Watson Research Center
provided the fundamental concepts of RISC, the idea came from the IBM 801
minicomputer built in 1971 which is used as a fast controller in a very large telephone
switching system. This chip contained many traits a later RISC chip should have: few
instructions, fix-sized instructions in a fixed format, execution on a single cycle of a
processor and a Load / Store architecture. These ideas were further refined and
articulated by a group at University Of California Berkeley led by David Patterson,
who coined the term "RISC". They realized that RISC promised higher performance,
less cost and faster design time. [1]. The simple load/store computers such as MIPS

1
are commonly called RISC architectures. David A. Patterson was the finder of the
term RISC, after that John L. Hennessy invented the MIPS architecture to represent
RISC [2].

The second section of this paper will provide a comparison between CISC and RISC.
In section 3, we present some related work showing the attempts that had been made
on these two architectures, based on different techniques. Section 4 presents the
CRISC technology. Our work will be focused on Intel's family processors to illustrate
the concept of CRISC. Finally, the last section will be the conclusion of this work.

2. RISC versus CISC

When designers create a new generation of processors,
improving performance if often the key goal. There are three main factors that affect
performance; they are [13]:

• How fast you can crank up the clock.

• How much work you can do per cycle.
• How many instructions you need to perform a task.

In this section, our comparison will be based on the above key factors, to highlight the
major differences between the two architectures CISC and RISC.

A CISC processor has most of the following properties: [1]

• Richer instruction set, some simple, some very complex

• Instructions generally take more than 1 clock to execute
• Instructions of a variable size
• Instructions interface with memory in multiple mechanisms with complex
addressing modes
• No pipelining
• Upward compatibility within a family
• Microcode control
• Work well with simpler compiler

As time passed, one of the non-RISC architecture with large market was the Intel x86
family, it has some specific characteristics that became associated with CISC:
[1]
• Segmented memory model
• Few registers
• Crappy floating point performance

Typically CISC chips have a large amount of different and complex instructions. It
was believed that hardware is always faster than software; therefore one should make
a powerful instruction set, which provides programmers with assembly instructions to
do a lot with short programs. Commonly speaking, CISC chips are relatively slow per
instruction compared to RISC chips, but use less instructions than RISC.

Most actual RISC machines such as the RISC I and RISC II from the University of
California at Berkeley and the MIPS from Stanford University have most of the
following common properties: [1]

2
 • Simple primitive instructions and addressing modes
• Instructions execute in one clock cycle
• Uniformed length instructions and fixed instruction format
• Instructions interface with memory via fixed mechanisms (load/store)
• Pipelining
• Instruction set is orthogonal (little overlapping of instruction functionality)
• Hardwired control
• Complexity pushed to the compiler

Additional properties of CISC and RISC, regarding cost and performance together
[14] are:
• RISC design is approximately twice as cost-effective as CISC.
• RISc architectures are designed for a good cost/performance, whereas CISC
architectures are designed for a good performance on slow memories.

Also, more properties added to a new RISC technology, including [16]:

• Superscalar and out-of-order execution

• Large number of registers
• Fast floating point performance
• Larger Cache.
• Higher bandwidth.

The essence of RISC architecture is that it allows the execution of more operations in
parallel and at a higher rate than possible with a CISC architecture employing similar
implementation complexity. It can not only improve parallelism using pipelining, but
also make superscalar and out-of -order execution possible [15].

RISC has also been called a "scalable architecture" because it is possible to go from
one technology to another with practically the same design [3].

Back in the middle to late 80's, the battle over RISC and CISC is mainly non-Intel
versus Intel x86, and RISC seemed to have a clearly upside, until the appearing of
i486, Pentium and now PII, PIII. Now Intel's machines still run the old instruction set,
but they adopt some RISC-like characteristics such as one clock execution, clean
memory models, deep pipelining, superscalar operations, lots of registers and even
out-of-order execution. They run faster and faster with a decent floating point
performance. On the other hand, some RISC machines added more instructions to
their architectures for new data types. So, it seems the RISC-CISC gap is narrowed.

So, nowadays, the difference between RISC and CISC is no longer one of instruction
sets, but of the whole chip architecture and system. The designations RISC and CISC
are no longer meaningful in the original sense. What counts in a real world is always
how fast a chip can execute the instructions it is given and how well it runs existing
software [1].

RISC's original goals was to limit the number of instructions on the chip so that each
could be allocated enough transistors to make it execute in one cycle. Rather than
provide a mul instruction, for example, the microprocessor's designers might make

3
sure that add executes in one clock. Then a compiler could multiply a and b by adding
a to itself b times or b to itself a times. A CISC could multiply 5 by 10 as follows:
[5]

Mov ax,10
Mov bx,5
Mul bx

But a RISC chip might do it like this:

Mov ax,0
Mov bx,10
Mov cx,5
Begin:
Add ax,bx
Loop Begin ; loop cx times

Table-1 below shows some examples of CISC and RISC processors:

CISC Processors RISC Processors

IBM 370/168 MIPS R2000
VAX 11/780 SUN SPARC
Microvax II INTEL i860
INTEL 80386 MOTOROLA 8800
INTEL 80286 POWERPC 601
Sun-3/75 IBM RS/6000
PDP-11 MIPS R4000

The two architectures, CISC and RISC, can be compared based on instruction set,
which is an important feature of computer architecture. The instruction set chosen for
a particular processor determines the way that machine language programs are
constructed. Another way for comparing these two processors is by studying the
available addressing modes. This point will give us an idea about the memory or
register referencing, which will be one of the important factors in performances
comparison. Other factors for comparing these two architectures can be the integer
and floating point units. Recently one of the most important factors is instruction
pipelining. The more pipelining stages that the processor has, the faster the processor
will execute the instructions. Several researchers have been working on instruction
pipelining, because of the impact of this feature on the overall performance. The
cache and the main memory were also primary factors affecting processor
performance. Microprocessor industries were always greedy for speed, while memory
industry were greedy for capacity, causing a big gab between CPU and memory
speeds [Car 2002]. The number of transistors in each processor can affect the speed of
the processor. If the processor contains more transistors, this means it will have more
gates. This makes the design of the processors at the gate level more compact, and as
a result it will be faster, and has a better performance.

4
Table-2 shows some system comparisons between selected CISC and RISC
processors: [4]

CISC examples RISC examples

IBM 370/168 VAX 11/780 88000 R4000
Year developed 1973 1978 1988 1991
The number of instruction 208 303 51 94
Instruction size (bytes) 2-6 2 - 57 4 4
Addressing modes 4 22 3 1
Number of GRP's 16 16 32 32
Cache size (KB) 64 64 16 128

3. Techniques used for evaluating CISC and RISC Architectures

3.1 Combining features of CISC processors and RISC processors:

A method has been proposed and implemented by Simon Garth to combine two
different processors with two different operating systems into one PC. There were
some hardware and software issues involved in that study. The CISC technology in
the PC was based on the Intel 80x86 processor, with DOS operating system. While
the RISC technology in the PC was based on the i860 processor, with UNIX operating
system [8].

Figure-1 shows the system architecture for the combined machine at high level of
abstraction.

The researcher concluded that the resulting ability to increase computation power by
adding extra hardware when software was added provides an effective means of
accommodating new, more demanding, applications within the confines of existing
machine [8].

3.2 Experimental comparisons between CISC processors and RISC processors:

Performance comparison between CISC and RISC processors using integer and
floating point benchmarks was a very popular technique for many years. A study done
by Gao, Tang, and Ding that compares MIPS R2000 instruction set, which represents
a pure RISC and 80386 instruction set, which represents a pure CISC. Based on some
experiments, the researchers obtained some meaningful statistical results.

5
Table-3 shows the summary of MIPS R2000 and 80386 architectures [1].

MIPS R2000 Intel 80386

Date announced 1986 1985
Instruction size(bits) 32 Variable
Address space(size, model) 32 bits, flat 32 bits, segmented with paging
support
Data alignment Aligned No
Data addressing modes 2 11
Protection Page Segmented Scheme
Integer registers (number, model, size) 31 GPR*32 bits 8 GPR*32 bits, 6 segment
registers*16 bits, 2 other * 16
bits
Separate floating-point registers 16*32 or 16*64 bits 8*80 bits
Floating-point format IEEE 754 single, double IEEE 754 single, double,
extended

The following conclusions have been drawn [1]:

1- CISC has richer Instruction set.
2- CISC has better code density, 80386’s average instruction length is less than MIPS
R2000’s.
3- 80386 has richer addressing modes, but in both integer and floating-point case, 3
addressing modes used for about 90%’s addressing, some addressing modes even
never used
4- For 80386, the operand types can be “Memory”, while in MIPS R2000, only
load/store will access memory, all operands are in registers.

Another experimental study has been done by Dileep and Douglas [9]. The study
compared an example implementation of the RISC and CISC architectures (a MIPS
M2000 and a VAX 8700) on nine of the ten SPEC benchmarks. The organizational
similarities of these machines provides an opportunity to examine the purely
architectural advantages of RISC. The RISC approach promises many advantages
over CISC architectures, including superior performance, design simplicity, rapid
development time [9]. The researcher's founding was that the RISC, MIPS M2000,
has significantly higher architecturally determined performance than the CISC, the
Digital VAX 8700.

3.3 The impact of system software on CISC and RISC processors:

Shifting from a CISC design strategy to a RISC design strategy isn't without its
problems. Software engineers should be aware of the key which arise when moving
code from a CISC processor to a RISC processor.

The performance of a RISC depends greatly on the code quality that it is executing. If
the programmer (or compilers) does a poor job of instruction scheduling, the
processor can spend quit a bit of time stalling, waiting for the result of one instruction
before it can proceed with a subsequent instruction.

Since CISC machine perform complex action with single instructions, where RISC
machines may require multiple instructions for the same action, code expansion can
be a problem.

In the nineties people at Sun developed the java language and also the Java Virtual
Machine, which contains a complete, stack-based, instruction list. It can be seen as a
CISC instruction list. But compared to RISC it is opposite in every detail:[Bli 2006]

6
The JVM byte code architecture is:

• Not reduced instruction set.

• Not register based.
• Not load-store (Can combine memory access and execution)

The instructions are:

• Not fixed-length.
• Not one per cycle.
• Not simple (Can do multiple memory accesses and/or multiple execution
cycles).

There have been several research papers that covered the impact of system software
on the processor and memory performance. John Ousterhout [10] evaluated several
hardware platforms (MIPS M2000, VAX 8800, Sun-3/75, and Sun-4/280) and several
operating systems platforms such as (Ultrix, SunOS, RISC/os, and Spriite). A set of
benchmarks was used in his research. The study highlight issues for both hardware
designers and operating systems people to think about [10].

The improvement of computing performance, without any hardware update, can be

achieved by either leaving optimization to the compiler, which todays recent ones are
pipelined (splitting a complex operation in to several simple operations), or to the
programmer [17].

4. Complex-Reduce Instruction Set Computers CRISC

4.1 CRISC the future of processor design
Intel's Pentium series chips are sometimes described as CRISC (Complex-Reduce
Instruction Set Computers), because they are hybrid between the two architectures.
This first started when Intel released its i486 processor, and it was widely repeated as
having a RISC integer unit [19].

The term, like its antonym RISC, has become less meaningful with the continued
evolution of both CISC and RISC designs and implementations. The first pipelined
"CISC" CPUs, such as 486s [19] from Intel, AMD, Cyrix, and IBM, certainly
supported every instruction that their predecessors did, but achieved high efficiency
only on a fairly simple x86 subset (resembling a non load/store "RISC" instruction
set).

In our study, we focused on the Intel family microprocessors, to illustrate the future
trend of CISC and RISC, which is CRISC. An example of CRISC is the the Intel
Pentium-Pro, which is an interesting blend of the two architectures. It still executes
the CISC instruction set, but the internal implementation is a high performance "Post
RISC" CPU [12].

7
Table-3 shows some information about the Intel Family [7].

Microprocessor (Intel Family) Number of Transistors Year of Development

4004 2300 1971
8008 3500 1972
8080 6000 1974
8086 29000 1978
8088 29000 1981
286 134000 1982
386 275000 1985
486 1.2 Million 1989
PI 3.1 Million 1993
PII 7.5 Million 1995
PIII 9.5 Million 1997
P4 42 Million 2000
Core 2 Duo 291 Million 2006

350000000
Number of Transistors

300000000
250000000
200000000
150000000
100000000
50000000
0
II

ue
6

6
6

I
04

08

80

86

88

PI

P4
PI
PI
28

38

48

D
40

80

80

80

80

2
e
or

Microprocessor
C

Number of Transistors

8
Table-4 shows additional information about the Intel Family [7].

Microprocessor (Intel Family) Speed Year of Development

4004 108 KHz 1971
8008 200 KHz 1972
8080 2 MHz 1974
8086 5 MHz 1978
8088 8 MHz 1981
286 12.5 MHz 1982
386 16 MHz 1985
486 33 MHz 1989
PI 66 MHz 1993
PII 233 MHz 1995
PIII 1.2 GHz 1997
P4 1.8 GHz 2000
P4 with HT. 3.06 GHz 2002
Pentium D 2.66 X 2 GHz 2005
Core 2 Duo 3.0 X 2 GHz 2006

7000000
6000000
5000000
Spe e d

4000000
3000000
2000000
1000000
0
II
PI

D
HT
P4
6
6
6

PI
04
08
80
86

88

PI

o
28
38
48

P
Du
40
80
80
80

80

P4

2
re
Co

Speed Processor

4.2 Architecture of Intel Core 2 Duo

In this subsection, we explain briefly the architectural aspects of the Intel's recent
processor, the Intel Core 2 Duo processor. The Core 2 brand refers to a range of
Intel's mobile, desktop dual- and quad-core 64-bit x86 CPUs based on the Intel Core
microarchitecture. The Core 2 brand for desktop, laptop, and workstation PCs was
introduced on July 27, 2006 comprising Duo (dual-core), Quad (quad-core), and
extreme (dual- or quad-core CPUs with higher speeds and unlocked multiplier)
branches [7].
Unlike the architecture of Pentium 4 or Pentium D branded processors, the Core
architecture does not stress extremely high clock speeds, but rather improvements in
the processor's usage of both available clock cycles and power [7]. This translated
into more efficient decoding stages, execution units, caches, and buses, etc, reducing
the Core 2 CPU's power consumption, while enhancing their processing capacity.
The design of Intel Core 2 Duo is chosen to maximize performance and minimize
power consumption. It emphasizes mainly on cache efficiency and does not stress on

9
the clock frequency for high power efficiency. Although clocking at a slower rate than
most of its competitors, shorter stages and wider issuing pipeline compensates the
performance with higher Inter-Process Communications (IPC’s). In addition, the Core
2 Duo processor has more ALU units [18].

Intel Core 2 Duo processor uses the following instruction sets: x86, MMX, SSE,
SSE2, SSE3, SSSE3, x86-64. The most resent instruction set that the Core 2 Duo
processor is using is Supplemental Streaming SIMD Extension 3 (SSSE3), Intel's
name for the SSE instruction set's fourth iteration. The previous state of the art was
SSE3, and Intel have added an S rather than increment the version number, as they
appear to consider it merely a revision of SSE3. SSSE3 contains 16 new discrete
instructions over SSE3 [7].

The Core 2 Duo processor has 14 pipelining stages. The slightly deeper pipeline
enables increased clock speeds and techniques such as memory disambiguation and
improved prefetch logic also help offset any advantage an integrated memory
controller offers. It includes the shared 4 MB L2 cache and the L1 cache with size 32
KB + 32 KB. Intel has chosen not to follow suit in their Core 2 processors, preferring
to use any additional on-die transistors to increase the Level 2 cache to 4MB, it has
291 million transistors [7]. The improvements implemented in the Intel Core 2 chips
make them unquestionably the more efficient processors, able to decode and process
more instructions per clock cycle.

Figure-2 The Block Diagram of Intel Core 2 Duo Processor

The five main features of Intel Core 2 Duo contributing towards its high performance
are [18]:
• Intel’s Wide Dynamic Execution
• Intel’s Advanced Digital Media Boost
• Intel’s Intelligent Power Capability

10
• Intel’s Advanced Smart Cache
• Intel’s Smart Memory Access

Table-5 shows additional information about two different Intel's Core processors
[7].

Core 2 Duo (Merom) Core Duo (Yonah)

Manufacturing Process 65nm 65nm
Die Size 143mm2 90mm2
Transistors 291 million 151 million
Clock Speeds 1.06GHz - 2.4GHz+ 1.20GHz - 2.33GHz
FSB Frequency 533MHz - 800MHz 533MHz - 667MHz
L1 Cache Size 32KB + 32KB 32KB + 32KB
L2 Cache Size 2MB - 4MB Shared 2MB Shared
Pipeline Stages 14 12
Decoders 1 complex + 3 simple 1 complex + 2 simple
Maximum Decode Rate 4+1 3
Reorder Buffer 96 80
SSE Units 3 1
Socket-M (PGA/BGA)
Socket Interface Socket-M (PGA/BGA)
Socket-P (PGA/BGA)

We believe that debate about the two architectures CISC and RISC is of no more
interest in today’s computer technology. Computers today, though they bear the brand
of RISC or CISC, are in fact not pure implementations of either one. They are truly
hybrid systems. RISC architects have adopted a larger set of instructions and CISC
architects have realized the benefits of implementing a core set of instructions that can
execute in a single CPU cycle. From our observation Intel's Core Duo processor has
improved the cache system. It has two high speed caches of size (L1 and L2) in each
core (processor, which will store most recently used main memory locations. It
typically requires only one to two processor cycles to access data as compared with 10
to 200 cycles for main memory access. We must confess that bigger caches will only
improve the hit rate, but decrease the access speed. The Intel's Core Duo processor
utilizes 14 stage pipelining, to speedup execution, although the data dependences will
remain one of the problems in processors using pipelining. Another fact in Intel's
Core Duo processor is the increasing number of transistors (291 Million), which
means closer circuits and as a result a better performance.

Most of the researchers, who have done some work in this field and focused on the
hardware or software aspects of the two architectures CISC and RISC, have, in fact,
chosen two old processors, one to represent a pure CISC and the other one to
represent a pure RISC. This type of research does not have much impact on the design
of the current chips. Secondly, the two architectures are not pure CISC and RISC
processors. The comparison between them will lead us to wrong conclusions, because
the recent processors are CRISC (Complex-Reduce Instruction Set Computer), a
hybrid processor of CISC and RISC features. These types of processors have been
focusing on the increasing number of pipelining stages and the number of instruction
sets.

11
 5. Conclusion
A new trend of CISC and RISC architectures is addressed. Some of previous works
was highlighted, and a new technology is presented, Intel’s Core 2 Duo processor. For
the best performance and scalability of the Intel Core 2 Duo processor, the following
are important factors: (1) fast cache-to-cache communication, (2) large L2 or shared
capacity, (3) fast L2 access delay, and (4) fair resource (cache) sharing.

References:
[1] Yi Gao, Shilang Tang, Zhangli Ding, "Comparison between CISC and RISC",
2000.
[2] John L. Hennessy, David A. Patterson, "Computer Architecture A Quantitative
Approach", Third Edition, 2006.
[3] Margarita Esponda, Ra'ul Rojas,"The RISC Concept – A Survey of
Implementations", 1991.
[4] William Stallings, "Computer Architecture Designing for Performance",
"Seventh Edition", 2006.
[5] Jeff Prosise, "RISc vs. CISC: The Real Story", PC Magazine, 1995.
[6] Carlos Carvalho,"The Gap between Processor and Memory Speeds", 2002.
[7] Intel Group, "Intel Core 2 Duo Processor Specification",
http://www.intel.com/core2duo , 2007.
[8] Simon C.J. Garth, "Combining RISC and CISc in PC System", 1991.
[9] Dileep Bhandarkar, Douglas W. Clark, "Performance from Architecture
Comparing a RISC and a CISC with Similar Hardware Design", 1991.
[10] John Ousterhout, "Why aren't Operating Systems Getting Faster As Fast As
Hardware", 1989.
[11] Stefan Blixt, "Processors Designs for Embedded Systems", 2006.
[12] Mark Brehob, Trvis Doom, Richard Enbody, "Beyond RISC – The Post RISC
Architecture", 1996.
[13] Marc Tremblay, "Challenges and Trends in Processor Design",1998.
[14] Jim Thomas, "Evaluation of Performance Instruction Set Architectures: RISC
versus CISC", 2003.
[15] Richard S. Piopho, William S. Wu , "A Comparison of RISC Architectures",
1989.
[16] Dileep Bhandarkar, "RISC versus CISC: A Tale of Two Chips", 1997.
[17] Carlos Jorge Lopes,"Improve Computing Performance, without any Hardware
Update", 2002.
[18] Tribuvan Kumar Prakash, "Performance Analysis of Intel Core 2 Duo
Processor",Thesis, 2007.
[19] Paul DeMone , “RISc vs CISC Stilll Matters”, 2000.

12

View publication stats