Hi,
I was wondering if it was possible to use CUDA-aware MPI to send data from a GPU to a CPU of a different rank?
For example, calling an MPI Send(a) with host_data use_device(a), and MPI Recv(b) where b is on the CPU?
- Ron
Hi,
I was wondering if it was possible to use CUDA-aware MPI to send data from a GPU to a CPU of a different rank?
For example, calling an MPI Send(a) with host_data use_device(a), and MPI Recv(b) where b is on the CPU?
Hi Ron,
I haven’t specifically tried this and I can’t find anything on the web that confirms that this should work.
However, I can’t think of any reason why it wouldn’t. Have you tried it?
-Mat
Hi,
I have not tried it yet, but I figure in the absence of cuda-aware mpi, it may be a more efficient way to collect data that the CPU needs to write out.
I may try it in the future.