GPUDirect / CUDA - MPI on Windows

Is GPUDirect a thing on windows? I can’t run MPI-Cuda code on Windows that worked just fine on the same box under Ubuntu.

Typically GPUDirect RDMA (what is often associated with MPI usage) is associated with Linux.

It should still be possible to use MPI on windows, but CUDA-enabled MPI that depends on GPUDirect RDMA (e.g. for direct transfers between a GPU and a IB card) probably won’t work.

That is rather unfortunate. Thank you very much for the quick response on this txbob. What Linux distro(s) would you recommend that I use? I have 2 GTX 1050 Ti’s that I was able to get to run on Ubuntu but their aggressive updates seemed to break my peripherals and some critical software all of the time. This last update I lost my desktop when upgrading from Ubuntu 17.04 to 17.10… Not ideal.

Follow the instructions in the linux install guide.

Personal preferences will play strongly into picking a Linux distro. I am partial to RHEL so would recommend looking into that.

My take on RHEL is that it is a stable and robust platform. Never any problems, with or without GPU in the picture. The drawback is that one may not get to play with the latest bleeding-edge technology (e.g latest toolchain), but I don’t need that. CentOS should be very close to RHEL, but I have never used it.