Hi, I recently tried to program with pycuda + CUDA8.0 + mpi4py,and I found that my desktop (with 1 i7 4770 and 1 GTX 1080)computing faster than the school’s HPC platform(with 2 Xeon E5-2660 v3 and 2 Tesla K80),We think this is because the performance of a single CPU core of E5-2660 v3 is not as good as i7 4770 and one GPU defaults to socket one CPU core. So,the question is how can I get an mpi process to call a GPU and multiple CPU cores?