Hi, 
there was this issue raised some time ago:
https://www.mail-archive.com/slurm-users@lists.schedmd.com/msg10799.html

We're experiencing exactly the same issue now with GPU nodes in power saving, some (but not all) jobs doesn't start because of that, and it's annoying users - badly.
Anyone experienced something similiar? Or maybe Stefan has a solution?
(we're on version 24.05.5)

Best regards
Patryk Belzak.
--
HPC Administrator
Wroclaw Centre for Networking and Supercomputing