Quad-Core Microprocessors Craciun Stefan-Viorel

Quad-core microprocessor
specific architectures

Coordinating Teacher
-Radescu Radu-

Craciun Stefan-Viorel
441F

1
Quad-Core Microprocessors Craciun Stefan-Viorel

 Introduction – Brief History of Multi-Core CPUs

Multi-core processing has been at the heart of the computing

revolution for more than a decade. The transition began with server CPU
manufacturers adopting multi-core processor architectures to address
growing performance demands, combined with the dramatic increase in
power consumption of single core processors running at high frequencies.
Multi-core CPUs can operate at lower frequency, often consuming less
power, and completing work much faster by running tasks in parallel
compared to single core predecessors.
Over the years, multi-core technology transitioned to desktop CPU’s
and gaming consoles. Some consoles included as many as eight CPU
cores to enable immersive, complex, and realistic gaming environments.
Notebooks and laptop PCs then began to integrate multi-core CPUs to
further enable high performance computing environments for users on the
go.
And just recently, at the beginning of 2011, multi-core CPUs -
beginning with Tegra 2 - started shipping in volume in tablets and
smartphones, enabling support for higher quality mobile applications as
well as longer battery life. Examples include HD video playback, 3D
gaming, multi-tasking, and 3D interfaces which were already stretching the
capabilities of existing single core mobile processors.
When Tegra 2 and dual-core CPUs were initially being discussed –
well before device availability – they were assumed to be power hungry,
not useful for mobile devices, and largely dismissed. However, after Tegra
2 was launched, these concerns were proven untrue, and every other
industry leader followed suit with their own dual-core CPU product lines.
In order to further increase performance, extend battery life and stay
within mobile power budgets, mobile processors will continue to increase
the number of CPU cores.
Quad core consists of four connected processors that are capable of
communicating. This can be done on a single chip where the processors
are connected typically by a multibus. Alternatively, the multiprocessor

2
Quad-Core Microprocessors Craciun Stefan-Viorel

system can be in more than one chip, typically connected by some type of
bus, and each chip can then be a multiprocessor system. A third option is a
multiprocessor system working with more than one computer connected by
a network, in which each computer can contain more than one chip, and
each chip can contain more than one processor. Most modern
supercomputers are built this way. A parallel system is presented with more
than one task, known as threads. It is important to spread the workload
over the entire processor, keeping the difference in idle time as low as
possible. To do this, it is important to coordinate the work and workload
between the processors. Here, it is most important to consider whether or
not some processors are specialpurpose IP cores. To keep a system with
N processors effective, it has to work with N or more threads so that each
processor constantly has something to do. Furthermore, it is necessary for
the processors to be able to communicate with each other, usually via a
shared memory, where values that other processors can use are stored.
This introduces the new problem of thread safety. When thread safety is
violated, two processors (working threads) access the same value at the
same time. Consider the following code representation:
A=A+1
When two processors P1 and P2 execute this code, a number of
different out come may arise due to the real fact that the code will be split
into three parts.
L1: get A;
L2: add 1 to A;
L3: store A;
It could be that P1 will first execute L1, L2 and L3 and afterward P2
will execute L1, L2 and L3. It could also be that P1 will first execute L1
followed by P2 executing L1 and L2, giving another result. Therefore, some
methods for restricting access to shared resources are necessary. These
methods are known as Thread safety or synchronization. Moreover, it is
necessary for each processor to have some internal memory, where the
processor does not have to think about thread safety to speed up the
processor. As an example, each processor needs to have a private stack.

3
Quad-Core Microprocessors Craciun Stefan-Viorel

 Quad Core Architecture

The Quad processor architecture is completely designed after

implementing towards the embedded concurrent processor. The concurrent
processor focuses more on the interaction between tasks, correct
sequencing of the interactions or communication between the tasks, and
the coordination of access to their resources to input and output devices
are the key concerns during the design of concurrent computing system.
For this the concurrent components like SIMD array, mapping and
duplicate memories are added to each processor as shown in the figure
below.

 Design

The proposed quad core executes the multiple tasks. These multi
tasks may be implemented by the Height tree evaluation technique and
reordering of instruction Execution is necessary for this proposed
architecture. The Clock is the heart beat of any processor. The processor

4
Quad-Core Microprocessors Craciun Stefan-Viorel

executes one instruction within one clock period. The quad core is
responsible for many concurrent computing operations and it consists of
three main modules. These are Core processor, Quad RISC Processor, I/O
Ports. Generic RISC Processor are called scalar RISC Processor because
they are designed to issue one instruction per cycle, similar to the base
scalar processor. Four representatives RISC based processors from the
year 1990, the sun SPARC, Intel i860, Motorola M88100, and AMD
29000.Each processor use 32 bit instruction length. The instruction set
consists of 51 to 124 basic instructions. We consider these four processors
as generic scalar RISC, issuing essentially only these four processors as
scalar RISC, issuing essentially only one instruction per pipeline cycle.
Among the four scalars RISC Processor, we choose to examine the sun
SPARC and i860 architectures. The sun SPARC is derived from the
original Berkeley RISC design. The concept of parallel and distributed
computing Processor Core mainly consists of SIMD array, mapping, and
duplicate memories. Duplicate memories: The most important thing in
quad processor mapping is the boundaries and data flow of concurrent
processors. Duplicate memories are important and make a design flexible
and have a high throughput for both parallel and distributed strategies and
are too efficient for concurrent Architecture. SIMD array: This SIMD array
supports the multidimensional array of data. It allows the simultaneous use
of multiple processors for solving a task. An SIMD array is a synchronous
array of Processing Elements under the supervision of one control unit and
all P.E’s receive the same instruction broadcast from the control unit, but
operate on different multiple data sets from distinct data streams. It is
usually loads data into its duplicate memories before starting the
computation. All these are working together as a single processor that
involves both the parallel and distributed concurrent strategy.
RISC Processors (RISC): In the present work, the design of an 8-bit
data width Quad Reduced Instruction Set Computer (RISC) processor is
presented. It was developed with implementation efficiency and simplicity in
mind. It has a complete instruction set, program memories and data
memories, general purpose registers and a simple Arithmetical Logical Unit
(ALU) for basic operations. In this design, most of the instructions are
uniform length and similar format. Arithmetic operations are restricted to
CPU registers. The Instruction cycle consists of three stages namely

5
Quad-Core Microprocessors Craciun Stefan-Viorel

fetches, decode and execute. Many numbers of RISC processors give the
highest performance per unit area for parallel codes. A larger number of
RISC processor cores allow a fine-grained ability to perform dynamic
voltage scaling and power down. The RISC processor core with a simple
architecture is easier to design and functionally verify. This processor is an
economic element that is easy to shut down in the face of catastrophic
defects and easier to reconfigure in the face of large parametric variation.

 FPGA IMPLEMENTATION

Modern FPGAs are large enough to implement Quad-Processor

Systems-on-Chip. Commercial FPGA companies also provide system
design tools. To aid our design decisions, we devise a design methodology
adapted to the specific challenges we are facing. Our methodology for
deriving a digital logic Implementation of the required functionalities
encompasses three steps, each of which comprises a set of choices. The
criteria we employ to evaluate the final set of choices are complexity and
scalability. We define the complexity of a unit by two metrics: the maximum
operating frequency and resource usage. Lower complexity is better; that
is, the unit has a higher frequency and consumes less resource. We define
scalability as the ability of the unit to expand in a chosen dimension with a
minimal increase in complexity.
To determine complexity and scalability, we examine the FPGA
mapping of the design. In the first step, we start by choosing a particular
architectural technique that provides a certain subset of the required
functionalities. The second step considers the structural design choices of
each architectural technique in terms of the temporal and spatial
parallelism necessary to meet throughput and latency requirements. In the
third step, we take into consideration the logic-implementation-specific
choices driven by the constraints of the target FPGA chip. It is vital to note
that the choices in all three steps are interdependent. For instance, an
initially appealing architectural technique may require logic design choices
that lead toward a prohibitively expensive implementation. In this case, the
choice of the architectural technique has to be reexamined. In case there

6
Quad-Core Microprocessors Craciun Stefan-Viorel

are feasible implementations of the selected architectural technique, the

required performance can be achieved by balancing the tradeoff between
throughputs and operating frequency. Higher throughput necessitates
higher design complexity. On the other hand, to achieve high operating
frequency, the design complexity should be kept low. As the resulting
maximum frequency is highly influenced by the CAD tools that
automatically place and route the circuit.

7
Quad-Core Microprocessors Craciun Stefan-Viorel

8
Quad-Core Microprocessors Craciun Stefan-Viorel

 Real life examples and comparisons

NVIDIA’s Project Kal-El processor implements a novel new Variable

Symmetric Multiprocessing (vSMP) technology. Not previously disclosed
publicly, vSMP includes a fifth CPU core (the “Companion” core) built using
a special low power silicon process that executes tasks at low frequency for
active standby mode, music playback, and even video playback. The four
main “quad” cores are built in a standard silicon process to reach higher
frequencies, while consuming lower power than dual core solutions for
many tasks. All five CPU cores are identical ARM Cortex A9 CPUs, and are
individually enabled and disabled (via aggressive power gating) based on
the work load.
NVIDIA’s Project Kal-El vSMP architecture extends the benefits of
dual-core processors by providing:
 Lower power consumption
 Higher performance per watt
 Faster Web page load times
 Higher Performance for Demanding Applications
 Faster Multitasking
 Higher Quality Gaming

1. Lower Power Consumption, Higher Performance

per Watt

As described in the whitepaper titled “Variable SMP – A Multi-Core

CPU Architecture for Low Power and High Performance”, one of the
primary benefits of quad core CPUs is delivering lower power consumption
than dual core and single core CPUs.
A common misconception is that a multi-core CPU consumes more
power than a single core CPU and causes significant reduction in battery
life. On the contrary, due to variable symmetric multiprocessing, the main
quad core CPU architecture of Project Kal-El is more power efficient and
9
Quad-Core Microprocessors Craciun Stefan-Viorel

delivers higher performance per watt than competing single and dual core
processors.
In order to meet peak performance demands in a multitasking
environment, a single core CPU not only runs at higher clock frequencies
and voltages than a multi-core CPU, but also takes longer periods of time
to complete a given task. Multi-core CPUs are able to use symmetrical
multiprocessing and distribute workload across multiple CPU cores. Due to
workload sharing, each CPU core can run at lower frequency and voltage
to complete a multi-threaded task, or multiple tasks in a multi-tasking
scenario. Also, due to lower operating frequency and voltage, each core
consumes significantly lower power and offers much higher performance
per watt compared to single core CPUs.

2. Faster Web Page Load Times

Mobile devices that have a multi-core CPU with SMP support will be
able to deliver a fast desktop PC-style Web browsing experience. Modern
browsers such as Google Chrome and Mozilla Firefox are now multi-
threaded and capable of spawning several concurrent processes. Each
page tab in a Chrome browser is a separate process, and each process
manages its own set of threads. Both the processes and threads are highly
parallelizable. Figure 1 below shows the utilization of each core of a quad
core CPU system while browsing the popular New York Times Website.
From the figure it is seen that the browser uses all four cores of the quad
core CPU, and this parallel processing results in a Web browsing
experience that is much faster than on a dual core CPU-based mobile
device. Quad Core CPUs also deliver higher performance for browsers that
support tabbed browsing. Figure 2 below shows the CPU utilization across
the four CPU cores when multiple tabs are opened in a Web browser. In
fact, tabbed browsing uses significant CPU processing and as seen in the
figure, may even saturate all four cores of a quad core CPU.

10
Quad-Core Microprocessors Craciun Stefan-Viorel

CPU Usage % CPU0 CPU1 CPU2 CPU3

50%

Time (second)

Figure 2 CPU utilization on Quad core CPU system during Web browsing

CPU Usage % CPU0 CPU1 CPU2 CPU3

100%

50%

Time (second)

Figure 3 Quad Core CPU utilization during tabbed Web browsing

On a quad core CPU based system, the operating system will be able
to allocate multiple Web scripts across the four CPU cores and deliver
much faster execution of JavaScript heavy pages. Results from Moonbat, a
Web based JavaScript benchmark, show that a quad core CPU delivers
almost fifty percent faster Web browsing performance compared to dual
core CPU based mobile processors.

11
Quad-Core Microprocessors Craciun Stefan-Viorel

Figure 4 Quad Core Performance Advantage on Moonbat JavaScript benchmark

3. Higher Performance for Demanding Applications

A key benefit delivered by multi-core processors is the additional
performance that they provide for demanding applications and use cases.
The NVIDIA Tegra 2 processor with its dual core CPU delivered faster
performance for applications such as photo editing, video transcoding, Web
browsing, and multi-threaded gaming.
Quad core CPU-based mobile processors deliver even higher levels
of performance for these types of applications, and enable more compelling
applications never before seen on mobile devices. Examples include:
 High quality video editing
 Image processing
 Audio/video transcoding
 Physics simulations
 Numerous productivity apps
 Many forms of location-aware computing
 Facial recognition

12
Quad-Core Microprocessors Craciun Stefan-Viorel

 3D stereo games and applications

 Virus scans
 File compression
Results from Coremark, a popular mobile CPU benchmark, are a
strong indicator of performance for CPU intensive multimedia applications.
For instance, Coremark shows that quad core CPUs delivers almost 2x the
performance of dual core CPU based mobile processors and almost 4x the
performance of single core CPUs.

Figure 5 Coremark Multicore CPU Benchmark Results

Media applications such as Image processing, audio/video

transcoding, and file compression inherently have high degrees of
parallelism, and can take advantage of symmetrical multiprocessing and
multi-core CPUs. Image processing on a mobile device has become a very
common use case due to the inclusion of one or more cameras on most
mobile devices.

13
Quad-Core Microprocessors Craciun Stefan-Viorel

For example, Photaf 3D Panorama is a highly rated Android

application that enables users to automatically capture 3D panoramic
photos and stitch together the captured images to be viewed immediately.
The heavy image processing involved in detecting edges and stitching
together the images benefit greatly from the quad core processing
capabilities of Project Kal-El. Performance measurements show it is able to
process and display captured panoramic images almost two times faster
than on dual core CPU based mobile devices.

Figure 6 Performance speedup on Photaf 3D Panorama photo processing application

Linpack is a widely used CPU benchmark that gives a good measure

of the performance that a processor can deliver running CPU intensive
tasks such as media processing. Results from the multithreaded Linpack
benchmark show that quad core Project Kal-El delivers almost sixty percent
higher performance over an equivalent dual core processor. Actual
applications that are tuned for quad core processors will show even higher
performance gains on a quad core processor.

14
Quad-Core Microprocessors Craciun Stefan-Viorel

Figure 7 Performance Scaling on Multithreaded Linpack CPU benchmark

Media transcoding is another use case that benefits from

multiprocessing. Mobile users commonly capture and edit audio and video
files on their phones before sharing them with their friends and social
networks.
Handbrake, a popular video transcoding application, delivers
significantly faster transcoding on a quad core CPU system compared to a
dual core CPU based system. The figure below shows that a quad core
CPU delivers almost a sixty percent performance speedup for Handbrake
video transcoding tasks.

Figure 8 Quad core CPU Performance benefit on Handbrake Video Transcoding

15
Quad-Core Microprocessors Craciun Stefan-Viorel

As users begin to leverage the performance of quad core CPU-based

mobile devices and use these devices as their primary computing device,
developers will create more applications that leverage the power of quad
core processing. The examples discussed above are just a few use cases
that benefit. One can expect to see high-end applications for photo editing,
video transcoding, multi-point video conferencing, multi-threaded java
apps, video editing, and other scenarios become available for quad core
mobile devices.

4. Best Multitasking Performance

Another important and highly visible benefit of quad core processing

is superior multitasking that delivers high performance and uncompromised
responsiveness. A variety of factors contribute to poor responsiveness. For
example, a fully utilized CPU, unavailability of memory bandwidth, non-
optimized driver stacks, etc., all could cause poor performance and
unresponsiveness. Mobile device users typically have several applications
running concurrently. For example, it is not uncommon to see applications
such as Web browsers, streaming music, email syncs, social network
syncs, and news feeders running concurrently on a mobile device. Under
such heavy multitasking conditions, a single core CPU not only runs out of
processing power to service multiple tasks, it also has to run at peak
frequency to cope with the heavy workload. This results in poor
performance and power consumption.
On a quad core CPU based mobile device, the operating system can
dynamically allocate workloads to the appropriate CPU core based on
current CPU loads and task/activity priority. For example, if two CPU cores
are busy processing long tasks such as Android application updates, file
compression, and media processing the OS can immediately allocate
latency sensitive tasks like touch inputs to the available third or fourth core
to deliver faster responsiveness. When there are multiple tasks running at
the same time, queued up request from these tasks can be serviced faster
when there are multiple cores available to service these requests.

16
Quad-Core Microprocessors Craciun Stefan-Viorel

5. Higher Quality Gaming Experience

Gaming consoles and PCs today all use multi-core chips that support
SMP technology, and most console and PC games are coded to take
advantage of the multi-core and SMP capabilities of the hardware. Today,
most game engines such as Unreal 3.0, Id Tech 5 and Frostbite are
multithreaded, and the engines are increasingly migrating to task-
processing models where the “size” of each individual job is reduced while
the number of threads is increased. These threads are used for tasks such
as audio, collision detection, artificial intelligence, user input processing,
game strategy and network communications. All modern game platforms
support many threads, and this trend will continue to increase on future
platforms.
Due to the higher performance delivered by quad core processors,
advanced gaming features such as real-time physics and real-time texture
generation can now be used in mobile games to deliver significantly higher
graphics quality and more realistic gameplay experience.

Real-time Physics Delivering Additional Realism

When elements from the physical world such as collisions, wind,

water, gravity, motion, and more are included in the virtual world of a game,
the natural expectation is that these behave exactly the same way as they
would in the physical world. For example, collisions between two objects
should result in realistic outcomes based on the mass, speed, and
momentum of the colliding objects, cloth should interact with wind resulting
in realistic turbulent movement along its surface, and objects should react
to gravity and respond accordingly. These effects that are based on player
actions within the game and real word laws add an additional level of
realism to games making them more immersive and engaging. Most hig h
end PC and console games employ these physics base effects to deliver a
truly immersive and visceral gaming experience.

17
Quad-Core Microprocessors Craciun Stefan-Viorel

Implementing these special effects in a game is very performance-

intensive because all the physics calculations required to implement the
effects need to be done in real-time based on the real-time player actions
happening in the game. The introduction of quad core mobile processors
will now enable developers to use physics to create mobile games that are
not only graphically rich but also very immersive. As demonstrated in the
Glowball game demo5, the collision of the ball with the barrels, the
scattering of the barrels on the floor, the interaction between the ball and
the cloth curtains are realistic and result in non-repetitive outcomes that are
based on the speed of the ball, point of collision, and mass of the barrels.

Figure 9 Glowball Demo with Real-time Physics and Dynamic Lighting. Left side enabled, Right side disabled

On Kal-El, the processing tasks for dynamic lighting, physics, artificial

intelligence and other game related CPU processing is shared across the
four cores. Due to this load sharing, none of the cores are overloaded and
the processor still has additional processing headroom to handle
background tasks without compromising the user experience.
The graph below shows the performance speedup that the quad core
CPU based Project Kal-El platform provides for the Glowball demo over an
equivalent dual core CPU based platform. The significant performance
18
Quad-Core Microprocessors Craciun Stefan-Viorel

boost delivered by quad core CPU based mobile processors will bring us
closer to true console and PC class games for mobile devices.

Figure 10 Performance benefit from Quad Core mobile processor for advanced games

Real-time Dynamic Texture Generation

As mobile games increase in complexity and visual richness, the

corresponding file sizes of downloadable game content have also grown
tremendously. Often, the game file size is so big that the time taken to
download the game exceeds the fifteen minote refund time provided after a
purchase for users to claim a refund if they don’t like the game. This is not
only creating buyer frustration, but also is demotivating buyers from
purchasing advanced games that require large file downloads.
Real-time dynamic texture generation techniques allow game
developers to code games such that textures required in the game are

19
Quad-Core Microprocessors Craciun Stefan-Viorel

created in real-time based on the narrative and state of the game. Since
the textures are generated on the fly, they do not have to be provided at the
time of purchase. Using this technique, game developers can reduce game
file sizes by several orders of magnitude.

Figure 11 Scene variations in games, all dynamically generated on Kal-El’s 4 CPU cores

Another key benefit of dynamic texture generation is that the game

can be coded such that the environment in the game changes based on
player input and customization. For example, the player can be given the
option to change the weather in the game and based on the chosen
weather condition the associated textures that affect the scene can be
dynamically generated.Thus this technique helps deliver a very interactive
20
Quad-Core Microprocessors Craciun Stefan-Viorel

gaming experience without significantly increasing file size or coding

complexity in the game.
Real-time dynamic texture generation is however a very performance
intensive task and requires the performance of multi-core CPUs. The
Windmill dynamic texture generation demo created by Allegorithmic uses
dynamic texture generation to create in real-time a texture set that is over
300MB using only 900KB of base texture set. In other words, if a developer
were to create this game without using dynamic texture generation, the file
size of this game would be approximately 300MB and by using real-time
dynamic texture generation techniques, the developer can reduce the size
of his game to approximately 1MB. Generating this large texture set in real-
time requires significant CPU processing power that cannot be met by
single and dual core mobile processors. Real-time texture generation on
dual core mobile processor would overload the two processing cores and
result in a poor gaming experience. A quad core mobile processor would
not only be able to easily handle this task but also have sufficient head
room to handle other tasks in the background.
The introduction of quad core CPU based mobile devices will be an
inflection point for mobile gaming. The processing capabilities delivered by
quad core mobile processors will significantly accelerate development of
console and PC class games for mobile devices. Games running on quad
core mobile processors will deliver higher level of graphics quality and
realism that sets a new standard for mobile gaming experience.

 Conclusion

Desktop CPUs transitioned from single core CPU architecture to dual

and quad core architectures many years ago. But consumers started to see
real benefits of multi-core CPUs only a few years ago. This is because the
software ecosystem required to leverage the power of multi-core CPUs was
not immediately available for desktop PCs. Except for multitasking with
different applications on multiple cores, truly multi-threaded browsers and
applications that were capable of leveraging the multiple CPU cores only
became available in quantity a few years after the introduction multi-core
21
Quad-Core Microprocessors Craciun Stefan-Viorel

CPUs. Today’s desktop PCs are seeing the many benefits of multi-core
CPUs.

In the mobile space, the transition from single core CPUs to multi-
core CPUs has been much quicker. The mobile software ecosystem is
evolving from the work already done in the desktop space to enable
support for multi-core mobile CPUs. Widely used mobile software already
supports multitasking and multi-threading.
The Android operating system evolved from Linux, and therefore has
native support for multitasking and multi-threading. Recent releases of
Android 2.3 and Android 3.0/3.1/3.2 have added several features that
improve the operating system’s ability to leverage the processing power of
multi-core CPUs.
Mobile browsers such as Firefox and Webkit are based on their
desktop counterparts and therefore natively include support for multi-
threading. The new browser included with Android 3.0 supports not only
multi-threading, but also tabbed browsing. These browsers are able to
utilize the increased processing power offered by multi-core CPUs to
deliver a faster and better Web browsing experience.
Due to rapid growth in mobile gaming, developers are porting popular
PC and console game engines to the mobile environment. These game
engines were originally developed for multicore desktop PC platforms, and
therefore will leverage the multi-core CPUs in mobile processors to deliver
immediate benefits for mobile gamers. Quad core CPUs provide significant
processing power for game developers and will enable developers to
include advanced physics effects, artificial intelligence, collision
detection/avoidance, virtual texturing, better network playability, and more.
Quad core CPUs and variable SMP technology will enable mobile
devices to further push the performance envelope, and allow application
and game developers to deliver new mobile experiences, all while
extending battery life for the most popular use cases.

22
Quad-Core Microprocessors Craciun Stefan-Viorel

 References
 Design and Implementation of Quad Core Architecture Using FPGA -V.Prasanth,
K.Raja Sekhar ( V. PRASANTH et al. Int. Journal of Engineering Research and
Applications ISSN : 2248-9622, Vol. 3, Issue 5, Sep-Oct 2013, pp.745-749)
 NVIDIA Corporation Tegra White Paper
 Wikipedia
 John Shen Modern Processor Design: Fundamentals of Superscalar Processors,
1st Edition,McGraw-Hill Series in Electrical and Computer Engineering, 2015.

23