Hello, unfortunately imho Slurm's documentation is "scattered". What I mean is that a single subject is not defined in one place but described in several places. It is so for cons_tres.
Among others cons_tres allows for individual allocation of GPUs and other TRES, provides additional submit options to jobs, such as --gpus, allows for default CPU or memory allocation per GPU, etc. cons_res allowed only for cores and memory. This is at least my understading. I ran 19.X and now I'm running 22.05. See documents below and search for "cons_tres", there are multiple entries in each document. I remember searching in the whole manual dozens of times when searching for answers :) :/ And sometimes answers are not quite self-explanatory and confusing. https://slurm.schedmd.com/gres.html https://slurm.schedmd.com/slurm.conf.html https://slurm.schedmd.com/cons_res.html Kind regards -- W dniu 13.09.2022 o 23:04, Patrick Goetz pisze:
I think reading the documentation is making me more confused; maybe this has to do with version changes. My current slurm cluster is using version 17.xLooking at the man page for gres.conf (https://slurm.schedmd.com/gres.conf.html) I see this:NOTE: Slurm support for gres/[mps|shard] requires the use of the select/cons_tres plugin.On my current (inherited) Slurm cluster we have: SelectType=select/cons_resbut users are primarily using GPU resources, so I know Gres is working. Why then is select/cons_tres required?
-- Kamil Wilczek [https://keys.openpgp.org/] [6C4BE20A90A1DBFB3CBE2947A832BF5A491F9F2A] Laboratorium Komputerowe Wydział Matematyki, Informatyki i Mechaniki Uniwersytet Warszawski ul. Banacha 2 02-097 Warszawa Tel.: 22 55 44 392 https://www.mimuw.edu.pl https://www.uw.edu.pl
OpenPGP_signature
Description: OpenPGP digital signature
