Cuda best bang for the buck?

[peter]

Now that our I7's are really struggling with the modeling I am looking to run matlab and fortran under cuda to give a speed increase.
After some advice, as benchmarks tesla v normal high end gpu's are hard to find.
thanks.

 

[ltnmzh]

I remember the DP performance of 5XX series cards is better than 6XX series.

 

[Friss]

benchmarks tesla v normal high end gpu's are hard to find.
thanks.

If you still need a comparison, people in my research group are currently using GTX590s for large scale X-ray CT imaging (mostly in matlab, some C/#/++), apparently the (not small) company funding the research often does also. I can ask about the actual scale and speed if you haven't already bought something.

 

[{SAS}TB]

I would also be interested in this - a mate of mine is looking at putting together a high end (work) rig for rendering / 3D Studio Max for his company and considering Quadro 4000 or something like a GTX 580 / 590 etc

 

[peter]

We purchased the 690 to give it a go ( I know the speed is lower than the 580 but much less heat seemed worth it )
Just waiting for the matlab parallel processing toolbox to arrive and I will benchmark it against the 6 core i7 we have running.

 

[Friss]

Just had a quick chat with the guy in my group, said his 590 was getting about 80 GUPS, and that they find the limiting factor with consumer GPUs is normally the memory.

GUPS doesn't mean a lot to me, but a paper on his desk from last year showed a bit of a comparison between architectures for MRI work, 24 cores of XEON @ 3.33GHz was getting about 9 GUPS, an NVidia 285 about 10, and a quadro 2000 about 15.