I've found that this question arised two years ago: https://bugs.schedmd.com/show_bug.cgi?id=4717
And it's still unsolved :( -- Pavel Vashchenkov 02.03.2020 17:28, Pavel Vashchenkov пишет: > 28.02.2020 20:53, Renfro, Michael пишет: >> When I made similar queues, and only wanted my GPU jobs to use up to 8 cores >> per GPU, I set Cores=0-7 and 8-15 for each of the two GPU devices in >> gres.conf. Have you tried reducing those values to Cores=0 and Cores=20? > > Yes, I've tried to do it. Unfortunately, it does not work. > > Moreover, If I set limit of CPU cores on 'cpu'partition to any number > between 37 and 20 it still would not start the GPU job after the CPU > one, despite having un-allocated CPU cores. > > Another test was the following: I limited 'cpu'partition to 39 cores and > 'gpu' partition to 2 cores (MaxCPUsPerNode=39 and MaxCPUsPerNode=2) > respectively. When I start CPU-job with requirements ntasks-per-node=38 > one task less than limit, GPU-job starts without any problem > > It looks like I have to limit ntasks-per-node of my CPU-jobs to > MaxCPUsPerNode-1 or less. If I use all available CPU cores, GPU-job does > not start in 'gpu' partition on thesame nodes. >
