When we put our platforms to the test in SPEC's viewperf OpenGL benchmark, one we've used frequently in the past, things started to get interesting.
The most plausable explanation here is that the faster three platforms are being limited by the video card. Otherwise, the results observed in the slower PC133 platform don't make much sense at all.
DRV-07 turned in some interesting results as well. We're at a loss to explain the behavior here, and we even ran the benchmarks several times to make sure there was not an error. It is almost as though the combination of both high memory and
FSB bandwidth were necessary for the scores to begin to improve drasically. Performance didn't scale with clock speed at all on most of the test setups, indicating a probable memory bottleneck. Only on the 500/333 platform did performance begin to improve at higher clock speeds.
Yet another rather interesting result. The DX-06 viewset seemed to exhibit a sort of 'plateau' type behavior. At a certain clock speed, performance would jump quite noticeably, but not scale very much higher beyond that point. Again, we are at a loss for a technical explanation on this one -- theories are of course welcome in our forums!
The remaining viewperf tests turned in somewhat more expected results, and so we haven't included them in the article body. They're available in the appendicies instead.