[slurm-users] Configuring AMD GPUs
Oleg Sivokon
osivokon at nvidia.com
Mon Dec 18 14:55:27 UTC 2023
Hello list.
I'm reading documentation about gres types, specifically this part:
https://slurm.schedmd.com/gres.conf.html#OPT_File which seems to be relevant to NVidia GPUs, but, correct me if I'm wrong, not so much relevant to AMD GPUs. I couldn't find any information about AMD GPUs being exposed as files in the same or similar way to how NVidia's drivers do that.
Consider this configuration in gres.conf for NVidia GPUs:
NodeName=node001 Name=gpu Type=v100 Count=1 File=/dev/nvidia0 AutoDetect=NVML
what would the analogous configuration look like for AMD GPU?
When I try to simply not specify the "File", and set "Type" to glxrXXX and "AutoDetect" to RMSI, then slurmd ignores this resource (from memory, the log will contain a message saying something like "ignoring resource without file" -- sorry, I don't have a system with these GPUs ready at the moment).
Thanks!
More information about the slurm-users
mailing list