I spent a lot of time trying to find out why all CUDA projects I compile get zero output from cudaGetDeviceCount. Finally I found out: my NVS3100 (lenovo T510) is not working in new CUDA SDK! I was absolutely bewildered by the fact that BarsWF md5 bruteforcer works okay on my notebook (though rather slowly, but I’d like to get CUDA working for testing purposes), but no project I compile with 3.2 CUDA SDK version seem to work. After long time tweaking nvcc parameters I decided to replace cudart_32_32_16.dll with cudart.dll taken from BarsWF. And a miracle occured: it worked.
Now I have a question to NVIDIA CUDA writers: why is NVS3100 not supported in 3.2 CUDA SDK? I don’t think v4.0beta works either but that’s what I did not test thoroughly. If you need, I can attach cudart.dll taken from barswf but you can easily find it yourself at http://3.14.by/files/BarsWF_CUDA_x32_0b.zip
And yes, I did compile with -arch sm_11 option.
By the way, Nvidia Control Panel reports that there is CUDA.
Windows 7 Professional RU
CUDA SDK 3.2 (nvcc: Cuda compilation tools, release 3.2, V0.2.1221)
cudart_32_32_16 version 220.127.116.1120
cudart.dll (from BarsWF) version 18.104.22.1680, CUDA SDK ver. 2.3
If you need any additional info or have any suggestions on how to make my NVS3100 work with current CUDA sdk, it will be appreciated.
P.S. Sorry for my english.