Is it possible to access a given CUDA card from different processes (i.e. applications)?
What happens if 2 different applications call cudaSetDevice on the same device? Will the 2nd call fail?
Is it possible to access a given CUDA card from different processes (i.e. applications)?
What happens if 2 different applications call cudaSetDevice on the same device? Will the 2nd call fail?
It is possible.
Second application will not fail. Driver will maintain application-independent queue of kernels and they will be executed on device.