Hello,
I have noticed some strange behaviour with CUDA 4.0, my cuda application cannot run properly on my debug GPU… nothing work, I can compile it in Visual Studio 2010 Professional without errors or warnings but if I run my application nothing work!! If I use cudaDeviceProp in cudaGetDeviceProperties and read struct I have major and minor number set to negative number!! But if I compile same project in Visual Studio 2008 Professional with CUDA 3.2 all works properly and major and minor number are both correct!! Have someone there noticed this strange behaviour??
P.S.
Code for CUDA 3.2 and 4.0 is identical without changes.