Possibly extension of an earlier question.
My C++ MPI code currently runs on an earlier port of MPICH on windows namely MPICH 1.4.1p. In past few months I have embedded lot of CUDA 6.5 C code in my MPI application but all cuda code is written only for a single node. Right now I want to write CUDA aware MPI code for multiple nodes, but as I understand MPICH 1.4.1p doesn't support it
So now I am contemplating to migrate my MPI-CUDA application to a newer MPI implementation which is CUDA aware. As I checked, only Intel MPI and Microsoft MPI are supported on windows. Does anyone know if these MPI implementations are CUDA aware ?
Aucun commentaire:
Enregistrer un commentaire