No CUDA-capable device is detected

Question

I am new to CUDA programming, and I am completely stuck. When I try to run either the supplied deviceQuery utility or one of the sample applications, I get the following error:

cudaGetDeviceCount returned 38
no CUDA-capable device is detected.

I am using Windows 7, 64-bit. I installed Visual Studio 2012, and then CUDA 5.0 64 bit. My graphics card is an NVIDIA GeForce GTS 250. Today I went to the NVIDIA site and re-installed the most recent drivers for the card.

I can't find any support for this error in Windows beyond "ensure you have a working driver installed". Does anyone have an idea of what could be wrong?

Answer 1

The problem as discussed in the comments is that you are trying to run a CUDA session on a machine that you are connecting to via RDP. This won't work with a WDDM device. Please refer to this question/answer for more discussion about it and suggestions for possible work-arounds. Unfortunately TCC mode won't be an option for you since you're using a GeForce device, but if you can switch to VNC or some other remoting method, you may have good results.

EDIT:

Note that using some magic which I cannot explain, NVIDIA has recently introduced new windows GPU drivers which may (should) work around the above limitation. Using recent NVIDIA drivers for your GPU should now be an alternative to enable usage of the GPU even in an RDP session. This capability would only apply to recent CUDA GPUs for which there is recent driver support (i.e. 353.xx or newer, r352 and newer). So cc2.0 GPUs should be supported, but driver support for cc1.x GPUs ended some time ago, so these older GPUs will not be able to take advantage of this capability.