multiple GPUs with Accelerate

Im not sure if Accelerate in Fortran 10 support multiple GPUs automatically.

In both cases, are there any technical resources that I can references?


Hi Tuan,

Currently NVIDIA only allow a single host thread/process to access a single GPU at a time. So in order to use multiple GPUs, you need to add a higher level of parallelization (OpenMP, MPI, pthreads, etc.).

Things aren’t yet as seem less with OpenMP as we’ed like. So you will need to distribute the work to each thread. Eventually we’ll have the compiler do this for you.

Hope this helps,