CORRECTED: CPU vs GPU bottleneck: some hard(-ish) data

[tfm 169]

Sorry for the double post: I screwed up with the attachments in the earlier one.For insomniacs:The attached JPEGS show the results of four benchmarks with an E8500 at stock speed and overclocked to 4GHz. My GPU is an 8800 Ultra superclocked by EVGA. The benchmark is the FSXMark07 download from AVSIM's file library.What (I think) they show is this:1. At the benchmark's standard "Global high" settings:(a) the average framerate increases almost linearly - or better than linearly - with the extra speed of the CPU; and(:( the % of time with at least 25fps and 30fps also goes up significantly: ie, smoothness is much better.2. However, if I use my own more onerous settings (shown as "Tim's settings" in the JPEGS: see * below for a detailed description), the picture is different. Overall framerates are lower, obviously, but also:(a) although the average framerate increases with the faster CPU, the increase is nowhere near proportionate to the extra speed of the CPU; and(:( on the other hand smoothness (ie, % of time spent with higher framerates) does seem to improve roughly in proportion with the extra speed of the CPU.Looking just at the differences between minimum FPS, average FPS and maximum FPSs, it looks as if something other than the CPU is holding back better performance under my higher, more demanding, settings.But if one concentrates instead on the % of time spent at >24fps and at >29fps, one finds that there is a pretty much direct relationship with CPU speed.I incline to the view that these results point to a GPU bottleneck once you expose the CPU to a dense graphic environment. But looking at the %s, there is still a noticeable "smoothness" dividend in using a faster CPU.Tim* The main points in "Tim's settings" are: maximum realism, no tips, maximum special effects, all sliders on the "scenery" tab to the right, except that Autogen is on "normal"; clouds - full detail and maximum distance (by the way this makes a BIG difference in cloudy skies); no AI traffic of any kind; no lens flare or ground shadows or light bloom).

[tfm 169]

Further apologies: I managed to get the graph right but I still posted the wrong table.Just in case anyone's actually interested, I attach, finally, the correct data table. Tim

[DescendDescend 25]

You are not being ignored. I am just tuning up my similar system. Still struggling to get the memory to run at the design speed. Your results are very interesting.

[tfm 169]

I don't mind being ignored - honest! I'm just sorry I can't seem to cut & paste properly.Tim