LOSTCIRCUITS

Speaking of performance, one other issue with reviewing professional graphics cards is the unhealthy lack of adequate benchmarking tools. Whereas in the gaming world an abundance of benchmarks is available, regardless of how flawed or not they may be, there is preciously little out there that might provide an estimate of the real performance of professional graphics adapters. While it is true that the Standard Performance Evaluation Corporation (SPEC) is regularly updating the industry norm benchmark "viewperf", it only takes a cursory look at the revisions of the different software packages employed to figure out that some of the benchmark versions are behind the current versions distributed by the software vendors. This is true for for example Maya which uses the traces implemented in the release versions up to 6.0 whereas the newer traces that yield substantially higher performance with the appropriate hardware are scheduled for viewperf 9.0 that is due out sometime later this year. A somewhat different scenario is the case in 3dsmax. Essentially, Autodesk, the publisher of 3dsmax and, as of recent, also of Maya have been pushing the development of DirectX in their applications but not updated the OpenGL traces for several revisions. The result is that the OpenGL performance lags behind what it could potentially be. Other issues arising from older software revisions are, amongst other things, that symmetric multiprocessing is not supported, which puts all processing burdens on a single CPU or, by extension, a single core. As a result, the graphics adapters are dramatically underutilized in most of the benchmarks since the CPU becomes the one big bottleneck.

The 4500 moniker identifies the current top model in the Quadro Lineup.

Likewise, specialty drivers as for example nVidia's Maxtreme plugin are not recognized. Given that the Maxtreme drivers can only play their forte if the graphics cards are really challenged, the roughly 70% performance increase with the side dish of correct recognition of for example the overshoot flag may not weigh in that much. As we were told, amongst the issues resposible for the lack of support are matters of intellectual property. In a nutshell, in order to implement the driver functionality, the appication needs to understand how the drivers work and that little detail is not being disclosed. Regardless of the how and why, the bottom line is that the last two paragraphs pretty much summarize the dire straights of professional graphics benchmarking.

Specialty application benchmarks are available as well (SPECapc) for a variety of benchmarks. However, the benchmark versions severely lag behind the release versions, including the extra handicap of being specifically written for one or the other version and, thus, history repeats itself. To add insult to injury, the personal learning editions, as nice as they are, have (for example in the case of Alias Wavefront's Maya) multithreading and SMP disabled, which, once again, makes the CPU the bottleneck for any reasonable high-end benchmarking.

Needless to say that acquisition of the respective full versions just for the purpose of using them as benchmarks is not an economic solution and getting freebees is quite impossible. There are notable exceptions, we have been able to get through to Autodesk at some point to get a full version of 3dsmax and recently, we got our hands on a full version of Lightwave 8.5, however, particularly the latter is not a good tool for OpenGL Benchmarking.

Next Page:    => The Devil in the Drivers =>