# of available cuda devices less than total # of GPUs?

@ptrblck Is there a chance this is an issue in my code, or do you think this is most likely an issue in the NVIDIA drivers?