Select the Appropriate GPU for Your CAD Workflow — and Your Budget, Part 223 Jul, 2014 By: Alex Herrera
Understand how power consumption, support for multiple monitors, and other factors can impact your choice.
Editor’s note:Read Part 1 of this article here.
When you begin shopping for a workstation-caliber graphics processing unit (GPU), having a price range in mind will narrow the options quite a bit. But it still invariably leads to perhaps the most intimidating issue in this whole shopping process: How to assess and compare the performance of the products in that price range.
Spec sheets may seem like the answer, but most overwhelm the reader with floods of numbers, and use metrics that may not provide a good basis for comparison. If the performance metrics you’ll find in marketing brochures are of interest to you, by all means check them out — but remember to take each with a few grains of salt. While they can hint at a rough performance level, they are generally too subjective in definition and too specific in usage to indicate whether the GPU can succeed in making your job more productive.
Ultimately, getting at least a rough handle on how a GPU might perform while running your type of workload requires some type of benchmarking. Although they are often misused and misinterpreted, benchmarks can be useful tools, and should factor into your GPU selection process. (Learn more about GPU benchmarks in “How Fast Is It? Assess Your Graphics Hardware.”) But that’s not all the information you’ll need.
Capabilities and Features That Benchmarks Don't Measure
Benchmarks are inherently limited to yielding approximations of performance for the things they measure — and there are many professional GPU capabilities that they don't measure at all. For example, the extent of a card's support for advanced display functionality, GPGPU (general-purpose computing on graphics processing units), and memory size and speed can make all the difference in productivity, yet won't show up in published results for popular professional benchmarks such as SPECviewperf.
Display real estate. The secret’s long out that increasing screen “real estate” — the size, resolution, and number of displays — is the single best way to improve your productivity. Most users who have switched to two or more displays will tell you they could never go back to one. Today, the typical professional GPU can manage two or more high-resolution displays on its own. However, lower-end cards are more likely to limit the resolution per screen, an extra consideration should multiple high-resolution displays be a concern for you.
High-end and specialty video support. The capabilities of today's entry-class GPU cards are impressive; still, there are limits at the lower end when it comes to video and display support. These limitations will push higher-demand users to consider more capable cards.
Consider 10-bit color, for example. Design and engineering often go hand-in-hand with product styling, making 10-bit support nice to have, if not mandatory, for the desired visual impact and accuracy for client pitches, previews, or commercial advertisement.
GPGPU. We've discussed nothing but rendering, video, and display so far. But remember, there's another use of GPUs that today's professionals should be considering in their shopping criteria: GPGPU. In its evolution from technological curiosity to mature, workstation-caliber tool, GPU computing has quickened its development pace in recent years. Although GPGPU usage for mainstream consumer or corporate applications is still evolving and looking for compelling footholds, the same isn't true for professional computing spaces.
Common tasks such as raytracing, financial instrument modeling, and engineering simulation — for example, finite element analysis (FEA) and computational fluid dynamics (CFD) — exhibit the type of computing characteristics that make them excellent candidates for GPGPU acceleration: lots of repetitive, highly parallelizable floating-point arithmetic. And that opportunity has not been lost on top CAD rendering and engineering simulation software vendors, with top applications including ANSYS (FEA and CFD), Autodesk Moldflow (CFD), Bunkspeed, and Dassault Systèmes CATIA (both photorealistic raytraced rendering) tapping GPUs to reduce wait times considerably.
If GPGPU acceleration is on your radar, make sure your card supports the application(s) in your workflow that could benefit the most. NVIDIA's Quadro and Tesla cards support the CUDA platform as well as OpenCL, while AMD's FirePro line supports OpenCL.