HP Compaq Deskpro EN 6233, Compaq Deskpro EN 6266, Compaq Deskpro EN 6300, Compaq Deskpro EN 6300A, Compaq Deskpro EN 6300C White Paper

White Paper

October 2001

Prepared by:
Workstations Division Engineering
Compaq Computer Corporatio n

Comparison of Intel Pentium III
and Pentium 4 Processor

Contents

Introduction................................. 3

Comparison of Pentium 4 and
Pen tium III A rchitecture

Benefits....................................... 3

Case for Performance.................. 5

SYSmark 2001......................... 5

3D WinBench 2000 –

Processor Test......................... 6

Summary.....................................8

Additional Micro-Architecture

Detail .......................................... 9

Pen tium III............................... 9

Pentium 4.............................. 10

Impact of DirectX 8.0.............. 12

15WD-1101A-WWEN

Performance

Abstract: This white paper summarizes key technology
adva ncements in the Intel® Pentium® 4 process or compared wit h
the previous-generation Intel Pentium III processor. Common
benchmark workloads are discussed to provide an illustration of
whic h areas of c omputing will benefit the most f rom th is new
ar c hitecture. Results of Compaq benchma rk testing, c omparing
results for both processors, are included to demonstrate the
performance gains realizable with the new processor.

Comparison of Intel Pentium III and Pentium 4 Processor Performance White Paper 2

Notice

The information in this publication is subject to change without notice and is provided “AS IS” WITHOUT
WARRANTY OF ANY KIND. THE ENTIRE RISK ARISING OUT OF THE USE OF THIS
INFORMATION REMAINS WITH RECIPIENT. IN NO EVENT SHALL COMPAQ BE LIABLE FOR
ANY DIRECT, CONSEQUENT IA L, INCI DENT AL, SPECIAL, PUNITI VE , OR OTHER DAMAGES
WHATSOEVER (INCLUDING, WITHOUT LIMITAT I ON, DAMAGES FOR LOSS OF BUSINESS
PROFITS, BUSINESS INTERRUPTION, OR LOSS OF BUSINESS INFORMATION) , EVE N IF
COMPAQ HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH DAMAGES.

The limited warran ties for Compaq products ar e exclusively set forth in th e d ocu mentati on accompanying
such products. Nothing herein should be construed as constituting a further or additional warranty.

This publication does not constitute an endorsement of the product or products that were tested. The
configuration or configurations tested or described may or may not be the only available solution. This test
is not a determination of product quality or correctness, nor does it ensure compliance with any federal,
state o r lo cal require ments.

©2001 Compaq Information Technologies Group, L.P. Compaq, the Compaq logo, Evo and Deskpro are
trademarks of Compa q Information Technologies Group, L.P. in the U.S. and/or oth er countries. Intel,
Pent iu m, and Celeron ar e trademark s of Intel Corpor ation in the U.S. and/or oth er cou n tries. Micr oso ft and
Wind ows are either registered tr ademarks or tr ademarks of Microsoft Corp or ation in th e United States
and/or other countries. All other product names mentioned herein may be trademarks of their respective
companies.

Comparison of Intel Pentium III and Pentium 4 Processor Performance
White P aper prepar ed by Workstations Divisi on En gineerin g

First Edition (October 2001)
Document Number 15WD-1101A-WWEN

Comparison of Intel Pentium III and Pentium 4 Processor Performance White Paper 3

Introduction

This White Paper provides information useful in understanding the differences between the
Intel® Pentium® 4 processor and the previ ous -generation Pentium III. A dis cus s i o n is inc luded
about the architectural differences in the two processors and the performance benefits they
provide. When evaluating performance, there is no single performance test (“benchmark”) that
can completely describe the performance of a complex system like modern microprocessor or
personal computer. It is important to obtain the complete performance picture. In other words, the
system should deliver high performance across the entire spectrum of applications such as
producti vity, multimedia, 3 D a nd Inter net. Each of the se app licat io n c ategories c aries a uniqu e set
of computation and data movement characteristics; thus it is important to realize how each class
of application would benefit or not from the new architecture. It is also important to realize the
investment protection delivered, where the new architecture will provide reasonable performance
gain for current applications while providing headroom for future growth as more and more ISVs
will fully take advantage of the new a rchite c ture. W ith that in mind, it is expected that there is a
non-uniform gain in performance, as each class of current application lends itself more to the new
architecture while others do not. Using the Compaq Deskpro EN platform equipped with 1 GHz
Pentium III processor as the baseline, benchmark results of the new Compaq Evo D500 platform
equipped with the 1.7 GHz Pentium 4 processor are presented as a comparison of the two
architectures.

Comparison of Pentium 4 and Pentium III Architecture
Benefits

As Internet and digital med ia become more p ervasi v e i n modern comput ing, the Pentium 4
processor is optimized for a new level of digital audio, video, photography and 3D performance.
For corporate users, the Pentium 4 offers excellent performance with added headroom for future
applications such as

• Java technology and XML, which will be increasingly enabled in Office XP, Windows® XP
and W eb services

• Enhanced 3D rendering for business analysis, video decompression for e-learning, and peer­to-peer interaction for improved collaboration

• Secure connections with support for latest encryption technology for data transfer and e­Commerce tr ansactions.

How are these potential enhancements possible with this new processor? Let’s explore the mic ro­archit ecture enhancements in th e P entium 4 processor:

Representing a breakthrough to a new level of computing, the Pentium 4 processor is a
completel y red esig ned v ers i on of t h e earlier Intel IA32 processor architecture or Pentium III
while maintaining backward compatibility with existing applications. This means the Pentium 4
processor protects user’s current investment in existing applications while providing new
optimized instructions, registers, and data structures for future applications.

Comparison of Intel Pentium III and Pentium 4 Processor Performance White Paper 4

The Pentium 4 proces sor is optimiz ed for la rge data sets transfer and handlin g. This means th e
customer will see significantly improved performance over previous generation Pentium III
processor s in applications that handle and require larg e amount s of data. This will ap ply to all
vert ic al app lications an d many hor izontal app lic ations , such as financial analysis applic atio ns,
where handling large dat a sets is the nor m. For a limited nu mber of horizontal ap plications, such
as Microsoft Word, performance is not enhanced and can even suffer, though differences are
typically made up for by faster processor speeds enabled by the new processor architecture.
Howe ver, it should be noted that the tren d in offi ce appl ications is f or grea ter a nd great er usa ge o f
graphics. Use of graphics presupposes the existence of data-intensive graphics-generation
applications, which benefit greatly (and noticeably to the user) from Pentium 4 enhancements.
Moreover, as noted above, the trend is also to increasing use of java technology and XML in
Office XP, Windows XP and Web services. Nevertheless, in the short term, if the customer’s
need is primarily for office applications and there is a budget constraint, Pentium III may still
offer an acceptable solution. However, the customer should be aware that Compaq expects that,
in the near future, office applications will be handling much more data requiring the architectural
adva ntages the Pe ntium 4 p ossesses.

Perhaps more important for the user is the fact that higher processor speeds from Intel will only
be available in the future in the Pentium 4. The Pentium III will offer no further increases in
processor speeds. (Intel will continue to r ef resh Celero n processors , however). This is illu strated
in Figure 1, which shows the roadmap for of Intel processor technology. This means that
regardles s of the a pplication, improvements in performance can onl y be obtained b y greater
processor speed available from Intel in the Pentium 4 processor.

Figure 1: Roadmap of Intel Processor Evolution

Comparison of Intel Pentium III and Pentium 4 Processor Performance White Paper 5

On t he surface, the a rchitecture of t h i s new class of Pentium 4 pr o cess o r looks t he same as the
Pentium III, but after one drills further down, the Pentium 4 is significantly enhanced to give
better levels of performance in terms of frequency and i nstru c tions execut ion per c lock. Thes e are
the two variables that measure the level of how fast an application executes and is defined in the
following performance equation:

Performance = M H z (Frequency) x Instr u ctions executed per clo ck (I P C )
The Pentium 4 processor addr es ses th e two variab les in the performance equatio n w i th the n e w

underlying silicon/logic implementation of what Intel calls NetBurst micro-architecture. The
NetBurst mi c ro-ar c hitecture more specifical ly atta c ks the frequency and IPC varia bles of th e
performance equatio
technology, its redesigned architecture of the complete instruction pipeline, its execution engine,
and its extension to the existing instruction set. As we move forward through this paper, the
benefits of this is will be more clearly explained

More detailed information can be found after the summary section.

QZLWKLWVDGYDQFHG PDQG PVKRUWO\DIWHUVLOLFRQSURFHVV

Case for Performance

Applications generally can be divided into two classes: 1) floating-p oint-based ap pli cat ions
that are memory- a nd bandw idth- intensive and, 2) integer-based a nd b asic office productivity
applications. Recalling the performance equation mentioned above, the IPCs achievable by the
above two classes of applic atio ns vary great ly due to t he var iation of branc hes in applica tion
code. This variation of br anches a f fects t he predic tability of co de flow. A higher probabilit y of
correct pr e dicti on yiel ds a high er potential IP C and, therefore, h igher p erfor mance. Floati ng­point-based multimedia applications tend to have branches that are very predictable and thus have
a higher IPC potent ial. As a result, these appl ic ations scale very well with f reque ncy and benefit
greatly from the new architecture of the Pentium 4. However, integer-based and basic office
productivity applications tend to have more random branches in application code, thus are more
difficult to predict. The result is less efficient use of the Pentium 4 architecture on these
applications. However, since Pentium 4 processors are available at higher frequencies than
Pent ium III, performanc e is still e n hance d accor ding to the perf ormance equa tion.

SYSmark 20 01

SYSmark2001 is a suite of application software and associated benchmark workloads developed
by Applications Performance Corporation (BAPCO). It is a tool that measures system
performance on popular business-oriented applications in the Microsoft Windows operation
system. SYSmar k c ontai ns twel ve (12) applica tion workl oads that are di vided i nto two
cat e gories: Offic e Productivity and I nternet Content Creation.

Comparison of Intel Pentium III and Pentium 4 Processor Performance White Paper 6

Figure 2: Comparison of Pentium III with Pentium 4 in SYSmark 2001 Benchmark Tests

Figure 2 cl early il l ustr ates the Pentium 4 perf ormanc e adva ntages over Pentiu m III. It is also cl ear
that performance gains in the Office Productivity workload are less dramatic when compared to
Internet Content Creation workload. In the Internet Content Creation workload, where the typical
workload is streamed in nature (Windows Media Encode for example), the application tends to
have branches that are very predictable resulting in performance that scales very well with
fre quency and ben e fits greatl y from t he new archit ecture of the P entium 4.

3D WinBench 2000 – Processor Test

3D WinBench 2000 measures system–lev e l 3D performance, i nclud ing CP U and gra phics
subsystem. To understand the processor 3D performance, this benchmark suite includes the
Pr oc essor Test which measures the CPU-intensiv e portion of the 3D gr aphics pipeline – geometry
and setup stage.

Comparison of Intel Pentium III and Pentium 4 Processor Performance White Paper 7

Figure 3: Comparison of Pentium III and Pentium 4 in 3D Winbench 2000 Processor Test

To display 3D obj ect s o n a 2 D comput er s creen, it is much easier t o r epr es ent 3 D objects as a
collection of polygons (usually triangles) than as curved surfaces. The larger the number of
triangles used to represent the 3D object, the more closely the approximation of the mathematical
description resembles the 3D object. The process of breaking up a 3D object into triangles is
called tessellation and involves an enormous number of floating-point vector calculations.
Objects in the real world have material properties and reflectivity and these impact how the
objects interact with light, the more lighting from various sources and angles, the more realism to
the object/scene. Again, ca lc ulati ons of light effects on 3D obj e c ts require la rge numbers of
complex floating-point vector calculations. The CPU index performance gain in the 3D
Winbench 2000 – Process or Test , bench ma rk, illustr ated in Figur e 3 , resulted fro m the increas e i n
floating- point performance of the Pentium 4 processor .

Comparison of Intel Pentium III and Pentium 4 Processor Performance White Paper 8

Summary

The Pentium 4 architecture offers significant innovations compared to earlier Pentium III
technology. These innovations lead to breakthroughs in performance that are measured and
substantiated by testing reported in this white paper.

The Pentium 4 proces sor is optimiz e d for large dat a sets transfer and han dling, so customers w ill
see significantly improved performance over previous generation Pentium III processors in
applications that hand le and re quire la rge amounts of data . Floatin g- point-based multimedia
applications tend to have branches that are very predictable and thus have a higher IPC
(Instructions executed Per Clock) potent ial. Integer - based a nd basi c of fice productivity
applications tend to have more random branches in application code, thus are more difficult to
predic t. This means t he IPC potentia l is not high, bu t the fact that P ent ium 4 is availab le in higher
frequencies than Pentium III results in increased performance with these applications.

It is important for the user to note the fact that higher processor speeds from Intel will only be
available in the future from the Pentium 4. The Pentium III will offer no further increases in
processor speeds. (I ntel will continue t o refr esh Cel eron processors, however). At some p oint,
regardles s of the a pplication, improvements in performance can onl y be obtained b y greater
processor speed. The customer should be aware that Compaq expects that, in the near future,
office applications will be handling a lot more data, thus resulting in the need for increased
processing power and efficiency that the Pentium 4 offers.

Comparison of Intel Pentium III and Pentium 4 Processor Performance White Paper 9

Additional Micro-Architecture Detail

Figures 4 and 5 provide an o verview of the micro -architectu res of the Pentium II I and Pentium 4
processors respectively.

Pentium III

Figure 4: Pentium III Micro-Architecture Overview

Comparison of Intel Pentium III and Pentium 4 Processor Performance White Paper 10

Pentium 4

Figure 5: Pentium 4 NetBurst Architecture Overview

Again, the NetBurst micro-architecture attacks the frequency and IPC variables of the
performance equation

ZLWKLWVDGYDQFHG PDQG PVKRUW

ly after) silicon process
technology, its redesigned architecture of the complete instruction pipeline, its execution engine,
and its extension to the existing instruction set, which is as follows:

• 20-Stage Pipeline as compared to a 10-stage Pipeline in the Pentium III – smaller workload

per stage but at significantly faster execution time

• Execution Trace Cache to remove the long latency associated with the instruction decoder

from the main executi o n loop i n the Pentium III

• Rapid Execution Engine where multiple Arithmetic Logic Units (ALUs) are executed twice

as fast as the core freque ncy, resulting in higher execution throughput, reduced exe c ution
late ncy, a nd extension of the total of execution por ts to seven (7) as compared to five (5 ) in
the Pentium III

• Advanced Transfer Cache with much higher throughput at 54.4GB/s for a 1.7 GHz Xeon (32

bytes x one transfer per clock x 1.7 GHz) to feed the data-hungry execution units as
compared to 16GB/s throughput at 1 GHz in the Pentium III

Comparison of Intel Pentium III and Pentium 4 Processor Performance White Paper 11

• Advanced Dynamic Execution with very wide windows of instructions (126 instructions

versus 42 instructions in the Pentium III) from which the execution units can choose to
execute, thus avoiding dependency stalls that would prevent execution units from doing
useful work. In addition, 4KB of branch target buffer (as compared to 1KB in the Pentium
III), and a multilevel advanced branch prediction algorithm to keep detail on the history of
past program branches, thus reducing by approximately 33% the mis-predictions rate as
compared to the Pent ium III.

• 400 MHz System Bus with enhancements to signaling scheme and bus protocols, thus

featuring data bandwidth and b us transfer ef f iciencies much higher tha n those o f the Pentium
III, as follows:

– 200% data bandwidth improvement (3.2GB/s (8 bytes x 400 Mtransfers/s) versus 1.06

GB/s (8 bytes x 133 Mtransfers/s))

– 17% latency improvement for first critical data read
– 46% latency improvement for 64-byte read
– 25% latency improvement for data write
– 64% latency improvement for 64-byte write
– New cycles every two clocks at 200 MHz versus every three clocks at 133 MHz
– 200% snoop bandwidth improvement (3.2GB/s (64 bytes/2 clocks @ 100 MHz) versus

1.06GB/s (32 bytes/4clocks @ 133 MHz)).

– Higher concurrent requests
– Faster interrupt servicing (bus message versus I/O cycles)

• Streaming Single Instruction Multiple Data Extension 2 (SSE2) with 144 new instructions

that deliver 128-bit SIMD integer arithmetic operation and 128-bit SIMD Double-Precision
Floating Point to reduce the number of instructions to complete a task or program, effectively
increasing IPCs.

Comparison of Intel Pentium III and Pentium 4 Processor Performance White Paper 12

Impact of Dir ectX 8.0

Opt imized us age of SSE/S SE2 ext e nsion a nd code flow optimization to take a dvantage of th e
new NetBurst micro-architecture, allow graphic drivers to make use of DirectX 8.0
programmable vertex and pixel shader s to pr oduce signifi c ant perfor ma nce gai ns as il lu stra ted in
Figure 6.

Figure 6: DirectX8 Performance Improvements