Hi,
The problem is that my launches (through cuLaunch) seem to always fail. I always get CUDA_ERROR_UNKNOWN. I have my code doing the launch implemented in a static link library that is linked into a DLL. I use the launching code from an executable that links with the DLL. Note that I have all my code implemented in the same DLL and all CUDA driver API calls are made in the same DLL so global variables used by CUDA should not be a problem. All other functions work fine (like cuMemcpyDtoH for instance).