When we test professional graphics cards, we first run SPECviewperf, which has become an industry standard for benchmarking top-performance graphics workstations. Modeling rather primitive operations, this test shows the geometrical performance of the graphics card in the OpenGL environment which is determined by the card’s hardware features as well as driver optimizations.
This test works by feeding the graphics driver predefined sequences of OpenGL instructions that are typical of specific professional applications. The scripts integrated into the test simulate a user working in projection windows of the following applications (the names of the corresponding subtests are given in brackets): 3ds max 3.1 (3dsmax-04), CATIA V5R12 (catia-02), EnSight (ensight-03), Maya 6.5 (maya-02), Pro/ENGINEER 2001 (proe-04), Solidworks 2004 (sw-02), UGS Teamcenter Visualization Mockup (tcvis-01) and UGS NX 3 (ugnx-01).
It must be noted that SPECviewperf results should not be viewed as highly accurate. Most of the applications simulated by this benchmark are outdated, and you may have completely different results in newer versions of the same applications. Besides, professional graphics cards have become so advanced that many SPECviewperf subtests are limited by the CPU rather than the graphics card as is indicated by the results of the 3dsmax-04 and ensight-03 subtests.
As for the other diagrams, it is easy to see that you must choose a professional graphics card basing on what applications you are going to run on it. For example, ATI’s FirePro series are far faster in Maya, VisMockup and UGS NX but Nvidia’s Quadro FX are ahead in CATIA, PRO/ENGINEER and Solidworks.
SPECviewperf offers one more interesting opportunity. It allows to measure the performance hit provoked by full-screen antialiasing (FSAA). The diagrams below show how the cards’ SPECviewperf results change when we turn FSAA on.
The Quadro FX 380 is not present here. This cheap professional graphics card could not ensure problem-free operation with FSAA. Interestingly, you cannot use FSAA higher than 8x on the AMD cards in the OpenGL environment.
We’ve got predictable results. It is the memory bandwidth that affects the card’s performance in FSAA modes. That’s why the graphics cards with wider memory interfaces have a smaller performance hit than their opponents with narrower memory buses.
SPECviewperf can give you some notion about the performance of professional solutions in OpenGL applications but it gives you no clue as to the cards’ speed in DirectX. However, the DirectX API has been getting popular on the market of CAD and CAM software. Microsoft’s API is already supported by 3ds max and AutoCAD, for example. And anyway, other real-life applications used by 3D designers and engineers use newer OpenGL libraries than those used by SPECviewperf. So, this benchmarking suite cannot replace tests in real-life programs. Let’s proceed to them now.