Catia Benchmark

Wolfgang Dautermann <[e-mail]>

Description

Catia Benchmark
Wolfgang Dautermann Michael Trzesniowski
November 23, 2006
1 General Information
1.1 Introduction
At FH Joanneum a benchmark for Catia V (currently R14, Windows version,
supportforotherCatiaversionsiscurrentlyunderdevelopment)hasbeenunder
development since December 2005. This benchmark was driven by the industry
with special cooperation by the AUDI AG.
From the hundreds of Catia functions only a small part had been selected,
and combined into a script without user intervention. As FH Joanneum has
been using CATIA as the standard tool in various lectures since 1999, the nec-
essary models could be selected from the rich pool created by students. The
functions tested include loading of models, graphical representation (including
rotation), 2D drafting, changing a parameter and updating the model, creating
an assembly out of CatParts, cutting planes in DMU and clash computation.
The results are written to a log le, and can be analyzed with an Excel sheet
that creates suitable graphs and arranges the various results into a number of
categories (aka tasks) - namely ”total time”, ” le load”, ” le close”, ”CPU ac-
tivity”and”CPU+graphicactivity”. Thegraphsproducedareshownin2ways
- once using absolute numbers, a second set showing the performance relative
to the BEST result (the fastest results is set to 100
The benchmark is available to the public, as are most of the models used.
The large assembly used for some tests is available on request. Intel GmbH
Germany ...

Sujets

• 1,1-dichloro-1-fluoroéthane
• Assembly Hall (Champaign)
• Computer Sciences Corporation
• Architecture 32 bits
• OpenMP
• Xeon Woodcrest

Informations

• Publié par: Mefog
• Nombre de lectures: 105
• Langue: English

Extrait

1

Catia Benchmark

WolfgangDautermann

Michael Trzesniowski

November23,2006

General Information

1.1 Introduction At FH Joanneum a benchmark for Catia V (currently R14, Windows version, support for other Catia versions is currently under development) has been under development since December 2005. This benchmark was driven by the industry with special cooperation by the AUDI AG. From the hundreds of Catia functions only a small part had been selected, and combined into a script without user intervention. As FH Joanneum has been using CATIA as the standard tool in various lectures since 1999, the nec- essary models could be selected from the rich pool created by students. The functions tested include loading of models, graphical representation (including rotation), 2D drafting, changing a parameter and updating the model, creating an assembly out of CatParts, cutting planes in DMU and clash computation. The results are written to a log file, and can be analyzed with an Excel sheet that creates suitable graphs and arranges the various results into a number of categories (aka tasks) - namely ”total time”, ”file load”, ”file close”, ”CPU ac- tivity” and ”CPU+graphic activity”. The graphs produced are shown in 2 ways - once using absolute numbers, a second set showing the performance relative to the BEST result (the fastest results is set to 100 The benchmark is available to the public, as are most of the models used. The large assembly used for some tests is available on request. Intel GmbH Germany provided a set of computers. The necessary software was installed on that computers, each computer was benchmarked with our Catia-benchmark and these are the results of our benchmarks.

1.2 Proceeding Each computer was rebooted before each benchmark. Programs such as virus scanners, screen savers, which would falsify the benchmark were (of course) disabled during the benchmark. The computers got a fresh windows installation before the benchmark (either supplied by Intel or by our central IT Services departement).

1.3 Simulation of additional load For the simulation of an addtional CAE application we used a small (OpenMP enabled) C-programm which does some mathematical calculations and used a

1

lot of memory - the sourcecode is available online.

1.4 Simulated tasks The currently used version of our benchmark consists of the following subtests: •drawing - load small Catparts and generate a 2D drafting view •Rotate - load small Catparts and rotate them •Open/Rotate/Close a complex assembly in Material view mode •Modify - load a moderately complex part, change some dimensions and update the part •Assembly - create an assembly out of parts

2

•load a pick-up disc, modify and update the drawing •DMU section cuts in a complex assembly •DMU clash computing in a complex assembly •Open/Rotate/Close a complex assembly in Shading view mode

SUN Ultra 40 – 2.6 GHz

Online information about the benchmarked machine: •Information on vendor homepage •Exact Windows XP System Information (not the preinstalled Windows XP64, but the Windows XP (32 Bit), we installed for the benchmark). •/proc/cpuinfoInformation about the system as reported by Linux (Knoppix 3.9): and /proc/lspci

2.1 Hardware/Software •two Dual Core AMD Opteron(tm) Processor 285 2,6 GHz •8 GB Ram •NVIDIA Quadro FX 3450/4000

•Microsoft Windows XP Professional Service Pack 2 •Catia V R14, Servicepack 5

2.2 Installation notes We had problems with our benchmark on the pre-installed Windows XP instal- lation - there was no VBA-Support in Catia available. After a re-installation of Windows XP 32 Bit everything worked fine. The assumption, that the reason for the problems was XP64 was proven wrong later (the benchmark worked on the Supermicro Woodcrest machine (XP64), but we were unable to find the exact reason and benchmarked the machine with XP 32bit.

2

2.3 Benchmark Logfiles These are the logfiles from our benchmarks, from which the graphics were gen- erated. To prevent caching effects, each computer was rebooted before each bench- mark test.

3

•1 run

•3 runs

•3 runs with an addtional CAE simulation program in the background (2 OpenMP threads)

•3 runs with an addtional CAE simulation program in the background (3 OpenMP threads)

•5 runs

•5 runs, the affinity for the Catia process was set to CPU 0.

Supermicro Woodcrest marked in 64 Bit mode

•Information on vendor homepage

–

3.0

GHz

–

Bench-

•Exact Windows XP System Information (preinstalled Windows XP64 Ver- sion).

•System information as reported by Linux: /proc/cpuinfo and lspci

3.1 Hardware/Software •2 Intel XEON Dualcore ”Woodcrest” CPUs (EM64T, 3,0 GHz)

•8 GB Ram

•NVIDIA Quadro FX 4500

•Microsoft Windows XP Professional x64 Edition Service Pack 1

•Catia V R14, Servicepack 5

3.2 Installation notes We expected, that the machine will cause problems, because the benchmark did not work with Windows XP 64 Bit on Sun, but it worked without problems. We also benchmarked the machine later in 32 bit mode, to examine, if there are differences in 32 and 64 bit mode.

3

3.3 Benchmark Logfiles - 64 Bit mode These are the logfiles from our benchmarks, from which the graphics were gen- erated. To prevent caching effects, each computer was rebooted before each bench- mark test.

4

•1 run

•3 runs

•3 runs with an addtional CAE simulation program in the background (2 OpenMP threads)

•3 runs with an addtional CAE simulation program in the background (3 OpenMP threads)

•5 runs, the affinity for the Catia process was set to CPU 2.

•5 runs, the affinity for the Catia process was set to CPU 0.

Supermicro Woodcrest marked in 32 Bit mode

•Information on vendor homepage

–

3.0

GHz

–

Bench-

•Exact Windows XP System Information (Windows XP 32 bit version).

•System information as reported by Linux: and lspci

4.1 Hardware/Software •2 Intel XEON Dualcore ”Woodcrest” CPUs (EM64T, 3,0 GHz)

•8 GB Ram

•NVIDIA Quadro FX 4500

•Microsoft Windows XP Professional Service Pack 2

•Catia V R14, Servicepack 5

4.2 Installation notes The Windows XP Professional x64 installation worked on that machine, but the machine was also benchmarked using Windows XP 32 bit, to show, if there are any diffences in 32 and 64 bit mode.

4

4.3 Benchmark Logfiles - 32 Bit mode These are the logfiles from our benchmarks, from which the graphics were gen- erated. To prevent caching effects, each computer was rebooted before each bench- mark test.

5

•1 run

•3 runs

•3 runs with an addtional CAE simulation program in the background (2 OpenMP threads)

•3 runs with an addtional CAE simulation program in the background (3 OpenMP threads)

•5 runs, the affinity for the Catia process was set to CPU 0.

Intel Whitebox

– Pentium D 2.8 GHz

•Information on vendor homepage

•Exact Windows XP System Information

5.1 Hardware/Software •2 Genuine Intel(R) Dual Core Pentium D CPU 2,8 GHz

•2 GB Ram

•NVIDIA Quadro FX 4500

•Microsoft Windows XP Professional Service Pack 2

•Catia V R14, Servicepack 5

5.2 Installation notes Nothing special. Installation of the requrired software caused no problems.

5.3 Benchmark Logfiles These are the logfiles from our benchmarks, from which the graphics were gen- erated. To prevent caching effects, each computer was rebooted before each bench- mark test.

•1 run

•3 runs

•3 runs with an addtional CAE simulation program in the background (2 OpenMP threads)

5

6

•3 runs with an addtional CAE simulation program in the background (3 OpenMP threads)

•5 runs

Supermicro XP32

– Clovertown 2.1 GHz

•Information on vendor homepage

•Exact Windows XP System Information

•and lspciSystem information as reported by Linux:

6.1 Hardware/Software •2 quad core Genuine Intel(R) ”Clovertown” CPUs 2,1 GHz

•8 GB Ram

•NVIDIA Quadro FX 3450/4000 SDI

•Microsoft Windows XP Professional Service Pack 2

•Catia V R14, Servicepack 5

6.2 Installation notes Nothing special. Installation of the requrired software caused no problems.

6.3 Benchmark Logfiles These are the logfiles from our benchmarks, from which the graphics were gen- erated. To prevent caching effects, each computer was rebooted before each bench- mark test.

•1 run

•3 runs

•3 runs with an addtional CAE simulation program in the background (2 OpenMP threads)

•3 runs with an addtional CAE simulation program in the background (3 OpenMP threads)

•3 runs with an addtional CAE simulation program in the background (7 OpenMP threads)

•5 runs

•5 runs, the affinity for the Catia process was set to CPU 0.

6

7

Supermicro XP32 – Clovertown 2.6 GHz

•Information on vendor homepage •Exact Windows XP System Information •and lspciSystem information as reported by Linux:

7.1 Hardware/Software •2 quad core Genuine Intel(R) ”Clovertown” CPUs 2,66 GHz •8 GB Ram •NVIDIA Quadro FX 3450/4000 SDI •Microsoft Windows XP Professional Service Pack 2 •Catia V R14, Servicepack 5

7.2 Installation notes Nothing special. Installation of the requrired software caused no problems.

7.3 Benchmark Logfiles These are the logfiles from our benchmarks, from which the graphics were gen- erated. To prevent caching effects, each computer was rebooted before each bench- mark test. •1 run •3 runs •3 runs with an addtional CAE simulation program in the background (2 OpenMP threads) •3 runs with an addtional CAE simulation program in the background (3 OpenMP threads), the affinity for the Catia process was set to CPU 0. •3 runs with an addtional CAE simulation program in the background (7 OpenMP threads) •5 runs, the affinity for the Catia process was set to CPU 0.

8

GraphicalresultsandSummary

8.1 Some graphical results Here you can see some graphical results of our benchmark.

8.2 Overview on all tests - all machines The following graphic displays the runtimes of all tests on all machines:

7

Figure 1: Overview on all tests

8.3 Task based analyses - 3 runs The following graphic displays the task based analyses from only 2 specific benchmark on all machines. We choose the results from 3 consecutive runs (av- erage time) with and without an OpenMP program running in the background.

Figure 2: Task based analyses - 3 runs

8

8.4 Task based analyses - all runs The following graphic displays the task based analyses from all benchmarks on all machines.

Figure 3: Task based analyses - 3 runs

8.5 Performance relative to the fastest machine The last picture shows the performance relative to the fastest machine:

Figure 4: Task based analyses - relative to best result

9

8.6 Summary Our benchmarks clearly shows Catia V R14 is a primarily CPU bound appli- cation and profits most from single threaded CPU speed. Modern operating systems like Windows XP allow to run parallel applications in the background - but even if enough free CPU-cores are available, the negative performance impact can be measured. The impact seems nevertheless to be small enough, to make the use of multi socket/cores in a modern CAE/CAD environment a reasonable investment. On Woodcrest we tested both the 32-bit and 64-bit versions of Windows XP Professionals. The runtimes of the benchmark were slightly different, but similar enough that a user won’t notice in day to day work.

9

Changes

2006-09-21: Benchmarked the first systems and created the homepage.

2006-11-09: Benchmarked the Supermicro XP32 (Clovertown 2.1 GHz) machine and added it to the homepage and whitepaper.

2006-11-14: Benchmarked the Supermicro XP32 (Clovertown 2.66 GHz) machine and added it to the homepage and whitepaper. Added more details about the benchmarked hardware.

2006-11-23: Added the new results to the whitepaper.

10