ERROR: Not enough GPUs on node

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi all,

I’m new to using CUDA and I’m having some trouble getting it running on my ROCKS 6.2 cluster.

I currently have 4 nodes in the cluster, each with a single GTX745.

When I run ‘run_linpack’ with mpiexec -np 4, I get the following error:

!!! ERROR: Not enough GPUs on node compute-0-0.local, 1 GPUs found, 4 GPUs required !!!

How can I configure the CUDA HPL to run using a single card on each PC? Am I using an incorrect version of HPL for my configuration?

Cheers!

2

The problem is that you don’t know how to use MPI.

You’ll need to learn how to use MPI to distribute your 4 processes to 4 different nodes in your cluster. This will involve a proper hostfile and other considerations. It has nothing to do with CUDA or HPL.