-
Hello all, I've been trying to run nekRS 21.1 on the HPC and I've been running into issues when scaling up to more then a single node. When I run the turbPipePeriodic case using 4 GPUs (on a single node) the case runs perfectly, but when I increase the number of GPUs to 8 (over two nodes) I get the following error.
Any help would be much appreciated. Thanks, |
Beta Was this translation helpful? Give feedback.
Answered by
stgeke
Dec 20, 2021
Replies: 1 comment 1 reply
-
By default nekRS maps each MPI tasks to a GPU (id is derived based on the local MPI rank). In your case it looks like the requested device IDs (0,1,2,3) are not available. |
Beta Was this translation helpful? Give feedback.
1 reply
Answer selected by
wlu00
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
By default nekRS maps each MPI tasks to a GPU (id is derived based on the local MPI rank). In your case it looks like the requested device IDs (0,1,2,3) are not available.