Originally Posted by cjb
Do you have the cuda 3.0 sdk installed? If so, what's the performance of the FDTD3D example? It reports MPoints/sec near the end. My gtx260m also reports what looks like the correct clock speed in nvidia-settings, but benchmarks are still slow.
Otherwise, some othe suitable benchmark would be a quick way to see if we're working.
I have to agree with you here. I use NAMD (a molecular dynamic program) and I was able to get one of the drivers to work a few months back, although I couldn't find the driver anymore (I tested all of them). After I updated the driver, I could not get the correct driver to work again. When it was working, I was able to run NAMD with speeds 1.5x higher than the 8-CPU server that we have here.
Then with the newer drivers, I was getting horrible speed, really a slow down of NAMD instead of speed ups (even a 1-cpu only without GPU is faster than CPU+GPU). With the latest beta (256.29), it improves a little (faster than a 1-cpu or even 2-cpu) but I can still run a 4-cpu of NAMD faster than cpu+GPU. While the 256.29 driver improves something (the frequencies), it **seems** like some other CUDA related (that I've no idea of) is still not optimal.