[slurm-users] Over-subscription for a GRES type

[Paul Browne]

Hello slurm-users,


This may be a silly question, but I was curious if the concept of over-subscription on a GRES has come up before, or is currently possible in recent SLURM releases?


We have a use-case in that the GRES being tracked on a particular partition are GPU cards, but aren't being used by applications that would require them exclusively (lightweight direct rendering rather than GP-GPU/CUDA workloads), and so some degree of over-subscription on the cards would actually be quite welcome to alleviate job pressure on that partition.


Many thanks,
Paul Browne
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.schedmd.com/pipermail/slurm-users/attachments/20181123/d6eafeff/attachment.html>