Originally Posted by RagingCain
He kept everything the same but went from tess off, tess norm, and tess extreme.
There is one variable you gentleman have forgotten, does tessellation bottleneck the GPU and the CPU? You guys would have to remove the CPU from the variable equation (high/stable overclocks.) The drivers still have to communicate with the CPU for the most part, have any of you guys tested increases CPU frequency to compensate for the extra levels of instructions having to be processed? Perhaps even adjusted inter-system communication such as QPI frequency / Uncore frequency. Boosting those at times can prove more effective then raw CPU speed.
I think CPU could become a factor in the no (or possibly the normal) tessellation tests if you're running high-end SLI and don't have at least a Bloomfield at 4.0GHz. In my experience Heaven will *start* to exhibit some signs of CPU BN at around 80fps, and will 'cap out' at around 150fps no matter how low I put the settings (iirc ... haven't tried in quite awhile).
Thus, I think if you were running like SLI 580's on a q6600, you'd end with with a very low perf hit from switching to extreme, because your 'no tess' run would be significantly dragged down due to the CPU speed. The extreme run ... much less so, due to the fact that you're shaving 30% off of the FPS, and hence lowering the probability of the CPU acting as limiting factor to performance.
However, so far nobody has turned in any wild outliers (like an uber-low 25% performance hit), so I don't think that (so far, at least), the CPU on anyone's rig is coming in to play to any substantial degree. Plus we've had people testing single-card vs. sli and coming up with the same % perf hit, which suggests that (at least on their rigs), the CPU isn't impacting the results.Edited by brettjv - 8/4/11 at 9:51am