CODEWITHSUNDEEP
openclslurmgpuopencl-c
R.Merritt
R.Merritt

Reputation: 477

Trying to get some OpenCL code to use 2 GPUs

Hi I have just inherited some OpenCL code that appears to only use 1 GPU I submit it using SLURM on a cluster and when I submit it requesting multiple nodes (typically 2) the program just runs twice. I am wondering where I go on how to structure / restructure the code so that it can take advantage of multiple GPUs.

Upvotes: 1

Views: 176

Answers (1)

VojtaK
VojtaK

Reputation: 678

So you have two different computers on the cluster, each has one GPU and you want to distribute one task between the GPUs on different machines?

I am not aware of anything that would directly support that. In OpenCL you won't create any runtime over two computers so that is out of the game as far as I know, you need to stick to one platform. I think the same apply for CUDA.

You can however experiment with the tools https://www.open-mpi.org/ or even https://github.com/alpaka-group/alpaka designed for distributing tasks over the cluster.

Upvotes: 0

Related Questions