[slurm-users] one MPS in one computing node.
wenxiaoll at 126.com
Mon Feb 28 01:52:30 UTC 2022
I am sorry for this email -- "MPS on A100 only zero-index GPU was used when there are four GPUs", wherein the https://slurm.schedmd.com/gres.html#MPS_Management says that SLURM is only supporting one MPS GPU on one computing node.
In Nvidia's official website, it says MPS can support multi cards in one Computing Node.
Are there any difficulties preventing us from achieving it or other reasons?
is there a plan to support multi-card MPS in one node in SLURM?
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the slurm-users