HP COMPAQ EVO W6000, COMPAQ EVO W8000 User Manual

White Paper

March 2002
16G1-0302A-WWEN

Analysis of Intel Xeon

Prepared by:
Workstations Division
Compaq Computer Corporation

Contents

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

Differences Between Intel

Xeon Processors ........................4

Intel Xeon .18 Processors......... 4

Intel Xeon .13 Processors......... 4

Benchmark Analyses ................. 4

Business Winstone.................... 5

SYSmark 2001.......................... 7

Cadalyst.................................... 8

ProE 2001i2 ............................ 10

Summary ................................... 13

Processor Frequency Grades
and Cache Sizes on
Performance Benchmarks

Abstract: Compaq Evo Workstations W6000 and W8000 refreshes
will feature the leading edge Intel Xeon .18 processor with Hyper­Threading Technology. The new processors are fabricated with the
latest .13µ technology, 512KB-L2 cache, the ability to support
frequencies ranging from 1.8 GHz to greater than 2.6 GHz and also
includes support for multi-threaded execution.

The purpose of this paper is to study the performance benefits in
terms of processor frequency grades and larger cache sizes of the
new Intel Xeon .18 processors versus the previous Intel Xeon .18
processors. This will be done using different industry-standard
benchmarks.

Analysis of Intel Xeon Processor Frequency Grades and Cache Sizes on Performance Benchmarks 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, CONSEQUENTIAL, INCIDENTAL, SPECIAL, PUNITIVE, OR OTHER DAMAGES
WHATSOEVER (INCLUDING, WITHOUT LIMITATION, DAMAGES FOR LOSS OF BUSINESS
PROFITS, BUSINESS INTERRUPTION, OR LOSS OF BUSINESS INFORM ATION), EVEN IF
COMPAQ HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH DAMAGES.

The limited warranties for Compaq products are exclusively set forth in the documentation 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 ens ure compliance with any federal,
state or local requirements.

Compaq and Evo are trademarks of Compaq Information Technologies Group, L.P. in the U.S. and/or other
countries.

Microsoft, Windows, and Windows NT are trademarks of Microsoft Corporation in the U.S. and/or other
countries.

Intel, Pentium, and Xeon are trademarks of Intel Corporation in the U.S. and/or other countries.
All other product names mentioned herein may be trademarks of their respective companies.
©2002 Compaq Information Technologies Group, L.P.
Analysis of Intel Xeon Processor Frequency Grades and Cache Sizes on Performance Benchmarks

White Paper prepared by Workstations Division
First Edition (March 2002)

Document Number 16G1-0302A-WWEN

16G1-0302A-WWEN

Analysis of Intel Xeon Processor Frequency Grades and Cache Sizes on Performance Benchmarks White Paper 3

Introduction

Computer performance is highly dependent on key system features. Processor speed, memory
bandwidth, graphics cards, and disk drives all play im porta nt rol es in dete rmining system
performance. Highly computationally intensive tasks can benefit from higher frequency
processors. A larger cache on the processors can reduce the number of memory accesses and
increase system performance on applications that have small data sets (those that can reside on
the processor cache). Applications that require very large files use many disk accesses and require
more optimization in that area. Memory bandwidth is a crucial factor in getting the data quickly
to the processor. This paper will focus mainly on system processor performance.

A good measure of performance is the amount of time it takes to execute a given application.
Contrary to popular belief, clock frequency (MHz) and the number of instructions executed per
clock (IPC) are not fair indexes of performance by themselves. True performance is a
combination of both clock frequency (MHz) and IPC.

Performance = Frequency x IPC

The formula: Performance = Frequency x IPC means that performance can be improved by
increasing frequency, IPC or both. Frequency is a function of both the manufacturing process and
the micro-architecture. At any given clock frequency, the IPC is a function of processor micro­architecture and the specific application being executed. Although it is not always feasible to
improve both the frequency and the IPC, increasing one and holding the other close to constant
with the prior generation provides a significantly higher level of performance.

In addition to these two methods for increasing performance, it is also possible to increase
performance by reducing the number of instructions that it takes to execute a specific task. Single
Instruction Multiple Data-Stream (SIMD) is a technique used to accomplish this. This is done
using 128-bit SIMD single-precision floating-point Streaming SIMD Extensions (SSE).

This analysis will discuss the performance differences between different speeds of the previous
Intel Xeon .18 processor (Foster) and new Intel Xeon .13 Processor (Prestonia). This paper will
also analyze how the larger cache on Xeon .13 determines system performance.

16G1-0302A-WWEN

Analysis of Intel Xeon Processor Frequency Grades and Cache Sizes on Performance Benchmarks White Paper 4

Differences Between Intel Xeon Processors

Intel Xeon .18 Processors

The Intel Xeon .18 processor (Foster) builds upon the Intel Netburst micro-architecture, built with
the 0.18-micron process and with 256KB L2 cache, which facilitates high-speed critical
calculations, memory accesses, and an Execution Trace Cache. The Execution Trace Cache
caches decoded x86 instructions (micro-ops), removing the latency associated with the instruction
decoder from the main execution loops. In addition, the Execution Trace Cache stores these
micro-ops in the path of program execution flow, where the results of branches in the code are
integrated into the same cache line. This increases the instruction flow from the cache and makes
better use of the overall cache storage space (12K micro-ops), since the cache no longer stores
instructions that are branched over and never executed. The result is a means to deliver a high
volume of instructions to the processor’s execution units and a reduction in the overall time
required to recover from branches that have been mispredicted. The trace cache is a micro­architectural design that has a direct impact in the Intel Pentium 4 (P4) core attaining a higher
IPC than the Intel Pentium 3 (P3). However, this has a drawback too. When the processor needs
to fetch a new instruction, it must rely on relatively much slower instruction decoders— thereby
causing the netburst architecture to idle and wait on the slow decoders.

The Level 2 Advanced Transfer Cache is 256KB in size and delivers a high data throughput
between the Level 2 cache and the processor core. The Advanced Transfer Cache consists of a
256-bit (32-byte) interface that transfers data on each core clock. As a result, the processor can
deliver a data transfer rate of core speed multiplied by 32 bytes, reported in GB/s. This
contributes to the processor's ability to keep the high-frequency execution units executing
instructions vs. sitting idle.

Intel Xeon .13 Processors

The new Intel Xeon .13 processor now features a 512KB L2 cache instead of the original 256KB
cache in the Xeon .18 Processor. The addition of the extra cache reduces the miss rates versus the
256KB cache misses. The size of the execution trace cache has not been changed nor have any of
the other units of the P4 core, but the increase in L2 cache will provide some performance
increase for most applications, especially newer ones. This will be evaluated in the following
sections.

The Xeon .13 processor is built with a 0.13-micron die shrink. The smaller transistors can switch
faster and produce less heat than their older counterparts. This can result in higher clock speeds
for these processors. All 0.13-micron CPUs use copper interconnects, which also aid in increasing
clock speeds.

Benchmark Analyses

These analyses are based on running benchmarks that focus on real-world applications run by
typical users running business appl ica ti ons, such as th e following:

• Microsoft Word

• Microsoft Outlook

• Users running Internet applica tio ns

16G1-0302A-WWEN

Analysis of Intel Xeon Processor Frequency Grades and Cache Sizes on Performance Benchmarks White Paper 5

• Users running 2D and 3D workstation applications on their computers
This analysis compares the same frequency Xeon .18 and Xeon .13 processors and two frequency

grades of Xeon .13 (2.0GHz and 2.2GHz) on a Compaq Evo Workstation W6000 system. Table 1
details the system configuration for the benchmarks.

Table 1. Benchmark Configuration

System Compaq Evo Workstation W6000

Number of Processors 1 CPU
Memory 512 RDRAM @ 800MHz
HDD 18GB 15K rpm SCSI 3 (U160 SCSI controller on board)
Graphics Card NVidia Quadro 2 Pro
Operating System Microsoft Windows 2000 Pro, SP2, 1-4 CPU
Hyper-Threading Disabled
Graphics Card Driver 12.95
Graphics Setting Vertical sync = Always OFF

Business Winstone

Business Winstone

Business Winstone is a system-level, application-based benchmark that measures the overall
peformance of a PC when running today’s top-selling Microsoft Windows-based 32-bit
applications on Microsoft Windows 98 SE, Windows NT 4.0 (SP6 or later), Windows 2000,
Windows Me, or Windows XP. Business Winstone runs real applications through a series of
scripted activities and measures the time a PC takes to complete those activities to produce its
performance scores. Higher scores mean better performance. When Business Winstone 2001 runs
the test, it runs at least a portion of each application through a script that was developed by
eTesting Labs. The script automatically executes commands within that application with no input
from the user.

Business Winstone 2001 uses the following applications in its tests:

• Norton Antivirus 2000 from Symantec

• WinZip 7.0

• Microsoft FrontPage 2000

• Lotus Notes R5

• Microsoft Access 2000

• Microsoft Excel 2000

• Microsoft PowerPoint 2000

• Microsoft Project 98

• Microsoft Word 2000

• Netscape Communicator 4.73

16G1-0302A-WWEN

Analysis of Intel Xeon Processor Frequency Grades and Cache Sizes on Performance Benchmarks White Paper 6

Content Creation Winstone

Content Creation Winstone is a system-level, application-based benchmark that measures the
overall performance of a PC when it is running under a 32-bit operating system, such as Windows
2000 or Windows XP.

Content Creation Winstone 2001 uses the following applications:

• Adobe Photoshop 5.5

• Adobe Premiere 5.1

• Macromedia Director 8.0

• Macromedia Dreamweaver 3.0

• Netscape Navigator 4.73

• Sonic Foundry Sound Forge 4.5

Following the lead of real users, Content Creation Winstone 2001 allows multiple applications to
be open concurrently and switches among those applications. Content Creation Winstone 2001 is
a single large test that runs the above applications through a series of scripted activities and
returns a single score. Those activities focus on "hot spots" or periods of activity when the PC is
working but the user is likely to only see an hourglass or a progress bar.

See Figure 1 for the Winstone Benchmark using Business Winstone and Content Creation
Winstone.

Winstone Benchmark

61.7

56.5

Higher is Better

Business Content Creation

65

2GHz Xeon .18

2GHz Xeon .13

2.2GHz Xeon .13

76.3

79.9

84.6

Figure 1

16G1-0302A-WWEN

Analysis of Intel Xeon Processor Frequency Grades and Cache Sizes on Performance Benchmarks White Paper 7

Analysis

Testing shows a 9% performance improvement using higher caches in the Xeon .13s versus the
Xeon .18 processor at the same frequency, when running business applications. The business
applications take advantage of the larger cache in Xeon .13s due to the reduction of memory
cycles, thereby bettering the performance. Only an additional 5% improvement is realized in
going to a higher frequency grade Xeon .13s.

In the Content Creation benchmark, a 6% increase is achieved in performance by moving to
higher frequency processors. However, only a 4.7% performance boost is realized from the
addition of 256KB of L2 cache in the Content Creation test, which could be caused by the disk
dependent nature of this test. Winstone Content Creation is not restricted by main memory
accesses to the same degree as Business Winstone. It puts a larger strain on the CPU core as is
evidenced by the better scaling observed when the core clock is increased. Both of these numbers
point to the fact that system benchmarks are not 100% dependent on CPU.

SYSmark 2001

Productivity performance should represent current business usage models of multi-tasking with
background computing. The Industry Standard productivity benchmark SYSmark 2001
incorporates mainstream applications for office productivity and Internet Content Creation, as
well as the latest business usage models to reflect platform productivity performance.

SYSmark 2001 is a suite of application software and associated benchmark workloads developed
by the Business Applications Performance Corporation (BAPCO), a non-profit consortium of
leading computer industry publ ica tio ns, indep endent testing labs, PC hardware manufacturers,
semiconductor manufacturers, and software publishers. SYSmark 2001 is a tool that measures
system performance on popular business-oriented applications in using the Windows operating
environment.

SYSmark 2001 contains fourteen application workloads that are divided into two categories:

• Office Productivity suite runs applications, including Dragon Naturally Speaking Preferred

Version 5, McAfee Virus Scan 5.13, Microsoft Access 2000, Microsoft Excel 2000,
Microsoft Outlook 2000, Microsoft PowerPoint 2000, Microsoft Word 2000, Netscape
Communicator 6.0 and WinZip 8.0.

• Internet Content Creation suite runs through Adobe Photoshop 6.0, Adobe Premiere 6.0,

Macromedia Dreamweaver 4, Macromedia Flash 5, and Microsoft Windows Media Encoder

7.

See Figure 2 for the SYSmark performance benchmark.

16G1-0302A-WWEN

Analysis of Intel Xeon Processor Frequency Grades and Cache Sizes on Performance Benchmarks White Paper 8

SYSmark Performance Benchmark

221

206

192

Higher is Better

Sysmark Rate Internet Content Creation Office Productivity

213

244

226

2GHz Xeo n .1 8
2GHz Xeo n .1 3

2.2GHz Xeon .13

187

173

201

Figure 2

Analysis

SYSmark 2001 is a horizontal benchmark that is a good measure of performance using a
multitasked load. This type of application has many branches in the code. BAPCO constructed
SYSmark 2001 to run in real time, meaning that the benchmark actually pauses for user input and
runs just as quickly as the tasks would if a normal user had been sitting at the keyboard
performing all of them.

The Office Productivity test is much more intense than Business Winstone and combines
conventional office tasks with virus scans and archive compression tasks, using WinZip.

The added cache results in an 8% boost for the Xeon .13 processor. An additional 7.5%
performance improvement is achieved using higher frequency Xeon .13.

The Internet Content Creation test is very memory bandwidth intensive since a large part of the
test is composed of encoding a video using Windows Media Encoder. A 6% performance boost is
achieved with the larger cache Xeon .13 processor and about an 8% boost in system performance
when going to a higher frequency Xeon .13 processor.

Overall, the additional L2 cache of Xeon .13 results in a 6% improvement in performance for the
Xeon .13 processor versus the Xeon .18. Additionally, about 6-8% performance boost is noted
with increased clock speeds in equivalent processors.

Cadalyst

For a more professional tool, the Cadalyst Labs C2001 benchmark for AutoCAD 2002 was used.
This test is comprised of two parts: 2D and 3D.

The 2D test opens drawings, runs array calculations, performs pan/zoom/erase /save operations,
and operations with orthogonal lines and cubical splines. The 3D test suite runs tests, which
include standard rotation/rendering, OpenGL rotation, DXFouts, in addition to some of the 2D
tests. Quadro2 Pro and Wildcat 5110 graphics cards were used to run this benchmark.

See Figure 3 for the Cadalyst performance benchmark.

16G1-0302A-WWEN

Analysis of Intel Xeon Processor Frequency Grades and Cache Sizes on Performance Benchmarks White Paper 9

Cadalyst on Quadro2Pro

26.17

24.96

23.21

Higher is Better

2D 3D

Cadalyst on Wildcat 5110

23.27

20.72

21.5

2GHz Xeon .18

2GHz Xeon .13

2.2GHz Xeon .13

30.14

2GHz Xeon .18

2GHz Xeon .13

2.2GHz Xeon .13

28.82

32.16

30.36

34.41

32.6

Higher is better

2D 3D

Figure 3

Analysis

In both 2D and 3D tests, the additional cache in Xeon .13 provides a 7% increase in performance
using the Quadro2 Pro graphics card and about 3-5% increase using the Wildcat 5110 graphics
card.

The 2D tests are less graphics card intensive and more CPU intensive, thus the CPU performance
greatly influences AutoCAD 2002 performance. From these scores alone, the 2.2GHz Xeon .13
shows about a 4% increase in performance from their 2GHz counterparts using the Quadro2 Pro
card and about an 8% increase using the Wildcat 5110 card.

In 3D tests, there is about a 7 % increase in performance going to higher speed Xeon .13s in both
cards. Quadro2 Pro performs almost 13-16% better than Wildcat 5110 in 2D tests and about 4-5%
better in 3D tests. This is because Quadro2 Pro drivers are more DirectX optimized. This points
to the importance of balancing system components versus focusing on raw CPU speed only.

16G1-0302A-WWEN

Analysis of Intel Xeon Processor Frequency Grades and Cache Sizes on Performance Benchmarks White Paper 10

In the next section, the performance of the system using more graphics-intensive tests using
OpenGL tests like ProE is examined by comparing the performance of the same processors with
two different graphics cards.

ProE 2001i2

The ProE 2001i2 benchmark is comprised of 17 tests. The model used in the benchmark is a
realistic rendering of a complete photocopy machine consisting of approximately 370,000
triangles. The ProE 16 graphics tests, each of which measures a different rendering mode or
feature, include the following, plus a time test:

• The first three graphics tests measure wire-frame performance using the entire model.

• The next four tests measure different aspects of shaded performance, using the same model.

Each of these tests executes exactly the same sequence of 3D transformations to provide a
direct comparison of differ ent rend er ing modes.

• The next four tests use a subassembly, and compare the two FASTHLR modes, the default
shading mode, and shading with edges. These tests also execute a common sequence of 3D
transformations.

• The last five graphics tests use two diffe rent instances of the model—the fi rst th re e with out
its outer skins (to illustrate the effect of FASTHLR and level-of-detail operations), and the
last two to illustrate complex lighting modes and surface curvature display.

• The last test is an aggregate of all time, not accounted for by the previous 16 tests, and is a
mix of CPU and graphics operations.

Scores are generated for all 17 tests. Composite numbers are provided for each set of graphics
tests (shaded, sub-assembly, wire-frame, and others) and there is an overall composite score for
graphics and CPU operations.

See Figure 4 for the ProE performance benchmark.

16G1-0302A-WWEN

Analysis of Intel Xeon Processor Frequency Grades and Cache Sizes on Performance Benchmarks White Paper 11

ProE- Difference between 2GHz Xeon .13 & Xeon .18

5.02

5.01

4.37

4.31
4

3.25

Higher is better

Composite Score Wireframe Composite Shaded Composite Sub Assembly

4.12

4.17

3.72

3.68

4.42

3.11 3.11

Figure 4

4.32

4.15

Composite

Wildcat 5110 Xeon-.18-2GHz

Wildcat 5110 Xeon .13-2GHz

Quadro2 Pro Xeon.18-2GHz

Quadro2Pro Xeon .13- 2GHz

4.51

4.49

4.07

3.33

Other Composite

4.1

Analysis

Figure 5 shows the results for the Intel 2GHz Xeon .18 and Xeon .13 using two different graphics
cards: WildCat 5110 and Quadro2 Pro. The extra cache in the Xeon .13 processor does not seem
to make any performance difference in using the Wildcat 5110 card. The Quadro2 Pro card shows
a lot more performance improvement using larger cache Xeon .13s. The Wildcat 5110 card
performs significantly better than the Quadro2 Pro card in almost all of the ProE benchmarks,
which leads to the conclusion that the graphics card performance is a lot more important in
determining the system performance in 3D applications, compared to processor performance.

Next, the test compared the performance between a 2GHz Xeon .13 and a 2.2GHz Xeon .13 using
the same two graphics cards.

16G1-0302A-WWEN

Analysis of Intel Xeon Processor Frequency Grades and Cache Sizes on Performance Benchmarks White Paper 12

ProE- Difference between 2GHz and 2.2GHz Xeon .13

Wildcat 2.0GHz

Wildcat 2.2GHz

Quadro2 Pro 2GHz

Quadro2 Pro2.2GHz

4.34

4.07

4.51

4.59

4.1

4.25

4.37

Higher is better

4.55

4

4.18

4.17

4.48

3.68

3.91

5.02

5.11

4.42

4.56

4.58

4.32

Composite Score Wireframe

Composite

Shaded

Composite

Sub Assembly

Composite

Other Composite

Figure 5

Analysis

Some performance improvement is achieved in moving to a higher frequency processor, but not
as significant as the performance difference between the two graphics controllers themselves. The
Wildcat 5110 card shows superior performance when compared to the Quadro2 Pro card, using
the same processor on the same system. Thus, the graphics controller plays a very important part
in determining system 3D performance.

16G1-0302A-WWEN

Analysis of Intel Xeon Processor Frequency Grades and Cache Sizes on Performance Benchmarks White Paper 13

Summary

Applications can be broadly divided into two categories:

• Integer/basic office productivity applications

• Floating-point/m ultimedia applications

The IPC cycle achieved by each of these different application categories varies significantly. This
is dependent on the number of branches that the application code typically takes, and the
predictability of these branches. The more difficult the branches are to predict, the higher the
possibility of mispredicting and performing nonproductive work, resulting in lower performance.

Integer and basic productivity applications, such as Microsoft Word and Microsoft Excel tend to
require several branches within the code that are more difficult to predict and, therefore, reduce
the overall IPC. As a result, performance increases on these applications take less advantage of
improvements in processor micro-architectures, such as deeper pipelines. This is seen by the 6­8% performance boost going to a higher frequency in SYSmark and Winstone benchmark results.
Additionally, smaller datasets can benefit more from the larger cache on the processor.

Floating-point and multimedia applications tend to have branches that are more predictable and,
therefore, contain a higher average IPC potential. As a result, these types of applications
generally scale very well with frequency and are inclined to benefit greatly from deeper pipelines.
Therefore, some performance improvement is seen in going to higher frequency processors using
Cadalyst and ProE benchmarks. However, there is not much performance improvement using
higher caches in these benchmarks. The graphics controller plays a more important role in
determining the performance of the system in running 3D benchmarks. Bigger cache and higher
frequencies offer some, but not a significant performance gain in 3D applications.

16G1-0302A-WWEN