Benchmarks: Time to Put your CAD Workstation to the Test
We showed you in an earlier post how to build your own customized CAD workstation from the ground up at an affordable cost, but even the best tool is only as good as what it is used for.
Now is the time to put your customized CAD workstation to the test and see what it can really do!
The day of reckoning is at hand - the money and toil poured into your CAD workstation must now face judgment. Sure, you might be sporting 2688 CUDA cores that can pump out 187 billion texels per second. A righteous sounding number, if you were paid by the texel that is. Spec dropping can quickly escalate out of hand, thanks to the internet. Who cares that your workstation leverages a tetra-bat of graphics to muster 70 billion more gigaflop-joules-per-fortnight-squared or that it might do the Kessel run in less than 14 – or perhaps 12 – parsecs? Before someone furrows their brow and disapprovingly folds their arms at your outlandish claims of CAD performance, it's high time to remove all doubt. Prove it.
We could subjectively argue ad infinitum about specific graphics card choices, API dependencies, and esoteric software limitations, but all truths are revealed by the right benchmark. For a benchmark to have value however, it must be representative of real-world experience, be repeatable, and must account for the interaction between hardware and software. Understandably, there's no one benchmark that does all these things perfectly. And no, checking to see if your workstation will run Crysis is not a relevant benchmark. Unless, of course, you want to play Crysis.
If it's so obvious that hardware specifications don't always translate to real-world performance why do they exist at all? Some of them are sadly just marketing ploys to create some imagined differentiation in a crowded market; for example whenever you see a display boasting a kajillion-to-one contrast ratio that uses an entirely unique -and therefore irrelevant- measurement methodology.
But not all hardware specifications arise from the seventh level of marketing hell. Hardware specs can indeed be useful as a normalized measure of hardware architecture. The aforementioned texel fill rate might be relevant in differentiating between two microarchitectures, say perhaps the GK104 in the Nvidia Quadro K4200 and the GM200-310 in a Geforce GTX 980 Ti. Will the fill rate be necessarily indicative of your installed performance in AutoCAD or Solidworks? Absolutely not. Hardware specs are an indication of potential, nothing more. Quite literally, YMMV. That's why we have recommendations.
There are myriad specifications to keep an eye out on, but especially relative to CAD you need to pay special attention to:
- single core CPU clock, cache
- process size
- microarchitecture version
- memory type
- memory bandwidth
- available memory (including cache sizes)
Similarly for the graphics subsystem, keep tabs on GPU microarchitecture, texel fill rates, core count and clock rate, dedicated memory type, memory bandwidth, and memory size. Again, you're really evaluating potential here so this only helps with your component selection, not benching an already built configuration.
By the way, CPU-Z is a great program to interrogate your system just to make sure you're really running what you think your running, and that some Jawa didn't try to push something with a bad motivator on you.
Synthetic benchmarks are custom code designed to tax your hardware in specific ways and generate results that can both be easily replicated and compared, often by generating a composite score. They are probably the most commonly used benchmark, best suited to validate whether your particular workstation configuration is reliably operating. After all, if your synthetic benchmarks are way off the mark for a known CPU/GPU combination, something may be seriously wrong with your hardware installation or driver configuration.
Best to resolve that before you start testing how you do with your favorite CAD software, if only to avoid embarrassment when your uber rig gets crushed by some five year old beige box and a hamster wheel just because you forgot to drop in the latest drivers.
Some useful synthetic benchmarks relevant for CAD workstations:
- SPECviewperf: If you use one benchmark for your CAD workstation, this is the gold standard for both OpenGL and DirectX GPU pipelines. Designed specifically to represent the behavior of typical CAD workloads through applet viewsets, and it even works on virtualized workstations. A more generic benchmark is available as SPECwpc.
- SiSoft Sandra: A robust set of synthetic benchmarks for CPU, GPU, and memory performance including power management and multi-core processor tests, also useful for interrogating your hardware.
- FurMark: An OpenGL graphics stress test.
- 3DMark: A synthetic benchmark focused on gaming, but relevant to general Direct3D performance
- HDTune: Synthetic benchmark to characterize your hard disk / SSD drives
- Super PI: Useful as a single-threaded CPU benchmark and stress test, with props to Japan and everyone's favorite irrational number. Relevant since most CAD packages, after all these years, still bottleneck on single-threaded CPU performance.
Real world benchmarks
The main drawback of synthetic benchmarks is that they indicate representative performance; they won't necessarily replicate the idiosyncrasies of your favorite CAD application. CAD software is notoriously complex and often highly temperamental with respect to certain hardware builds. Sometimes performance may be CPU rather than GPU limited.
This is especially true in the case of large assemblies and complex models where a single-threaded CPU process, struggling with the overhead of the editing software, has trouble feeding a relatively lightly loaded GPU process due to simple lighting/shading setups. The reverse might be true in a more GPU intensive activity, such as high resolution rendering.
As such, in the CAD universe, software is a critical part of the performance equation. That necessitates testing with production software. Fortunately there are a variety of real-world benchmark tests out in the wild that run scripted tests and provide results with the comparability of synthetic benchmarks:
- SPEC: While the SPECviewperf benchmark is an aggregate test across multiple workflows, the SPECapc benchmark families provide relevant (and long) tests for specific CAD setups including PTC Creo 3.0, Siemens NX 8.5, and SolidWorks 2015.
- Cadalyst Benchmark Test: Free download used to benchmark any basic version of AutoCAD released over the last sixteen years, exclusive of the industry specific AutoCAD implementations like Civil 3D or Mechanical.
- SolidWorks Performance Test: A Dassault provided benchmark specifically designed to share SolidWorks scores.
Ultimately, even the canned real world tests may prove to be insufficient, simply because your particular CAD workflow may be significantly different. That may very well mean you need to roll your own real world benchmark. Most CAD packages sport a scripting engine that when combined with suitable test files of whatever you typically work on, can allow you to homebrew your own workstation validation benchmark.
Time the results and use multiple runs to ensure consistency. Just make sure to use something representative of your largest jobs to properly explore your full design workload - that usually means the largest assemblies you have. Keeping a dedicated script and assembly for the test can allow you to consistently test changes moving forward, to validate the impact of specific upgrades, for example.
Just the facts
To settle internet arguments over hardware, there's nothing like benchmarks to quite literally settle the score. Maybe you can squeeze 11 parsecs out of that old piece of garbage you call a workstation. Use hardware specification for initial selection/sizing, synthetic benchmarks for configuration validation, and real-world benchmarks to prove software performance. Now go forth and post your results.
Share your own customized CAD workstation tips and tricks in the CAD/CAM/CAE forum on GrabCAD Groups.
We're making professional 3D printing easy and accessible
There's a new solution coming your way this summer. Click here or on the graphic to the left to request updates and beta access.
About the author: (Ed Lopategui)
Technology evangelist, entrepreneur, and aerospace engineer specialized in the software tools and technology which enable engineering, design, and product development - PDM, PLM, CAD, CAE, CAM. Any views, opinions, prophecies, and sarcastic remarks are my own and are in no way associated with any current or past employers.
All posts by Ed Lopategui — Follow on Twitter