Hi,
Are you just changing the drop down in the VC project settings for this? Which SM version are you building and what CUDA toolkit are you targeting?
One of the problem I had si similar to this one with GTX660Ti card. I installed Nsight 3.0 to work with 2008 VS. The sample CUDA projects default to CUDA 4.2 which is fine to debug from VS debug with break point in the host code. But when you put in the device/kernel code, it disabled/grey automatically. I tried it with CUDA 5.0 (both CUDA are installed), it behaves the same. I don’t have error message. By the way I setup to use local debug. According to User Guide, the Nsight should compile with -G0 option but my is always wit “-G -gencode=arch=compute_10,code="sm_10,compute_10.…” is this something related? Why is not -G0? I also asked/confirmed the right procedure of how to debug device code in my other forums but I will email you what I asked, thank you if you can take a look.
The correct option to enable debugging should be just “-G” instead of “-G0”. Older projects may still have it set to “-G0” in the CUDA project properties.