On the issue of lower OpenGL performance from the GTX series of cards, I did some research, and the conclusion seems to be that NVidia did some internal bandwidth bottlenecking on the GTX cards to keep them from competing with the much more expensive Quadro series. As gaming cards, the GTX series is designed to get images rendered to the screen quickly, but for OpenGL it is limited in getting those rendered images back to the CPU, which is what we need for graphics work. The CUDA pipes are wide open, from what I could find.
This stuff came up as part of my blog post exploring where the bottlenecks actually are in GPU accelerated rendering.
> Hi
>
> Just for fun i tried running my new 16core (32 threads) Z820 through some of these tests, and comparing it there also seems to be some odd numbers for your GPU in cineBench. I have the new Quadro K5000 that as far as i know is a lower clocked version of the GTX680 you have in your machine, but my OpenGL fps number are much higher. I only got the K5000 because i have had some bad experience with my former GTX card that just crashed the OpenGL driver when it was under load for longer periods. Would have gone with the GTX680 if i trusted the gamer cards more, but "known" stability was prioritised.
>
> CineBench scores:
> OpenGL 81.45 fps
> CPU 25.07 pts
>
> And running GeekBench (the free 32bit version) i can see that Teddy has some impressive memory numbers, higher than i get. Is the memory linked to the CPU clock frequency also? 4.5Ghz is a bit higher than my 3.1Ghz :-D
>
> Integer Processor integer performance 32414
> Floating Point Processor floating point performance 42123
> Memory Memory performance 7071
> Stream Memory bandwidth performance 8585
> Geekbench score: 28360
>
> not to show off, but i thought some of you guys would like to see the numbers of the current top of the line intel xeon
> well, but i also guess that this machine has cost ALOT more than what you spent on yours. Intel/HP sure know how to price their current highend stuff :-)
>
> - Robert
>