0

I am working on parallelising a code using MPI-2. It is successfully speeding up while I am using 8 core processors.

I was just wondering what would be the effect if I use GPUs for the same purpose instead of CPUs? According to my research so far, OpenMPI is compatible with GPUs and speed-up is only better if inter-processor communication is a lot as compared to CPU. Has anyone tried it out, or has any opinion on the same?

user1105630
  • 65
  • 1
  • 1
  • 5

1 Answers1

0

GPU parallelization is a totally different direction from MPI parallelization, in that GPU threads can't act like MPI ranks and make MPI calls - they're not really even threads in the sense needed. You can combine them, though, by calling GPU code in each rank of your MPI program.

Phil Miller
  • 36,389
  • 13
  • 67
  • 90