Table of Contents Overview
The purpose of this benchmark comparison is to evaluate the performance
of a Pro/Engineer (Pro/E) running on Windows NT graphics workstation
versus a Silicon Graphics workstation.
I created four benchmark tests and ran them on each type of SGI workstation and on our NT workstation.
The results were dramatic. In three out of four tests, the NT workstation outperformed the all of the SGI machines, including the Octane.
While the SGI machines have sharper graphics and a more refined
feel when running Pro/E; the price-to-performace ratio of the
$9,000 NTl workstation eclipses the $28,000 Octane machine.
Additionally, the NT workstation doubles as a superb business
software machine.
The
Contestants
Three different SGI workstations were compared to a custom built
NT Workstation. The hardware configurations are as follows:
|
|
Supplier: Corporate
Computer Systems, Cleveland, OH Processor: Dual Intel Pentium II 266Mhz Motherboard: SuperMicro P6NDH Graphics Board: Dynamic Pictures Oxygen 402 System Memory: 256MB 60ns EDO RAM Storage hardware: 4GB Seagate Ultrawide w/ Adaptec 2940 SCSI Controller Display: ViewSonic 21" G810 OS: Microsoft Windows NT Workstation 4.0, Service Pack 3.0 |
|
SGI Octane $28,000 |
Supplier: Rand
Corp. Processor: MIPS R10000 195Mhz IP30 Motherboard: Proprietary SGI Graphics Board: SI System Memory: 128MB RAM Storage hardware: 4GB UW SCSI Display: 19" SGI OS: IRIX 6.2 |
| SGI Solid Impact |
Supplier: Rand
Corp. Processor: MIPS R10000 195Mhz IP28 Motherboard: Proprietary SGI Graphics Board: Solid Impact System Memory: 128MB RAM Storage hardware: 4GB UW SCSI Display: 19" SGI OS: IRIX 6.2 |
| SGI Indigo2 |
Supplier: Rand
Corp. Processor: MIPS R4400 200Mhz IP22 Motherboard: Proprietary SGI Graphics Board: GR3-Elan System Memory: 128MB RAM Storage hardware: 4GB Wide SCSI Display: 19" SGI OS: IRIX 5.3 |
Method
of Testing
Pro/E allows commands to be run from script files called trail
files. Each time a user runs Pro/E, a trail file is created.
A trail file is essentially a log of all the commands that were
run during a work session, thus any work can be reproduced as
long as the trail file from that session is available.
To create an identical benchmark that can be run on any machine, a set of commands is run during a session of Pro/E and saved (automatically) as a trail file. This benchmark trail file is then run on different machines to measure performance.
Performance for a particular set of commands (i.e. shading and spinning a model) is determined by measuring the amount of time elapsed between the beginning and end of the command set. This done by inserting a time-stamp in the benchmark trail file with the MISC-TIME command in Pro/E.
After a benchmark is run, the trail file, created by running the benchmark trail file itself, is processed by a Perl script which extracts all of the time-stamps recorded during a session. The elapsed time for each set of commands is tallied and graphed.
I created four different benchmarks trail files that represent the ways Virtual uses Pro/E:
1) Large IGES Model using the Magna Delta Front Seat Assembly. This benchmark measured Pro/E performance using one of the largest IGES models (10-20MB) we have encountered at Virtual. General model manipulation (shade, spin, etc,) was measured as well as the time to create plot files.
2) Typical Pro/E Model w/ IGES MMP Gmx-210. This benchmark represents a typical automotive program. The assembly contains approximately twenty average-sized part files (100-500KB) and some medium sized IGES models (3-5MB). General model manipulation (shade, spin, etc,) was measured as well as the time to create plot files and the time to export IGES.
3) Large Pro/E Model -DN Pallet. This benchmark contains a very large number of average-sized part files in the main assembly. It is represents an uncommonly large and involved modeling project. General model manipulation (shade, spin, etc.,) was measured as well as the time to create plot files and the time to export IGES.
4) Computational model. This benchmark demonstrates
computational demands in Pro/E through a helical spring
model with several hundred windings. Since helical swept
features tend to severely tax the CPU, this model measures the
processor performance more than graphics performance. General
model manipulation (shade, spin, etc.,) was measured as well
as the time to create plot files and the time to export IGES.
Benchmark 1
Benchmark 2
Benchmark 3
Benchmark 4
Total of all benchmarks
The superior performance and low cost of the NT Workstation make it the clear choice for a Pro/E machine. Cost savings are realized not just in hardware prices, however. The NT Workstation runs all Intel software exceptionally well eliminating the need for additional PCs in the office for engineering use. Also, hardware maintenance agreements are unnecessary with the NT Workstations because hardware replacement costs are low. Finally, systems administration is simplified with NT Workstation resulting in definite time savings when compared to SGI administration.
Overall, significant long-term cost savings are possible by purchasing NT Workstations instead of SGI Workstations.
Performance of high-end PCs is expected to dramatically increase in the coming years. 700 MHz machines are due out by 1998; competition in the 3D graphics market should lower hardware prices; enhanced memory chips and faster bus speeds are expected in the near future as well.
PTC plans to support dual processing under NT with release 20 of Pro/E. Dynamic Pictures will release drivers for the Oxygen board that supports dual processors by the end of 1997 boosting OpenGL performance by a claimed 90%.
It should be noted that the NT Workstation is not as refined in terms of graphics as the SGI workstations. The graphics output of the NT Workstation, while very fast, is less crisp than the SGI box. (This is a subjective matter; personally, I did not notice a significant difference after using the NT box for an extended period). The NT Workstation, does support anti-aliasing which sharpens the graphics noticeably; and, unlike an SGI, graphics modes can be easily changed -- all benchmark tests were run in 1280x1024x16bit.
Robustness is a very important issue in machine comparison. Most complaints with NT workstations running Pro/E are about unreliable performance. The Dynamic Pictures Oxygen 402 is fully supported by PTC. I did encounter one problem, though: periodically when using Pro/E (one or twice a session) the a model would blank out while rotating, only to reappear when the rotation was finished. This problem appears to have been fixed in the latest driver release from Dynamic Pictures --- I have not observed the problem since updating the drivers. Other than this, the robustness of the NT Workstation is very good. No other problems with the operation of Pro/E have been observed. I used the NT Workstation extensively in a 200 feature modeling job for First Alert without incident.
A final issue to consider is the operation of Pro/Mechanica. Benchmark tests for Mechanica have not been conducted. Pro/Mechanica does run well on the NT Workstation however. Ryan Hopping has run medium sized FEA analyses (typical of an automotive mechanism design) without problems. Full benchmark testing is planned for on SGI Octane and NT Workstation.
It is my final recommendation that we evaluate the performance of large Mechanica models on the current NT Workstation and in lieu of no serious performance issues, begin migrating our engineering hardware to NT Workstations. I do not propose replacing our existing Indigo2s, rather, as new engineers join the company we should purchase NT Workstations instead of SGI's. This slow migration will allow us to take advantage of the benefits of the NT Workstations while minimizing the inherent risk in changing our engineering hardware.
All rights reserved. 1997 Virtual Engineering, Inc.