-gpu-mem option works only with --gpus=1

[Gldkslfmsd]

WRONG: the constraint list of gpuram sizes should be only 48g, not the others!

qsubmit -q gpu-* -gpu-mem 48g -c 8 --logdir=logs-faster "$x"

sbatch -o logs-faster/qsubmit.o%j -J qsubmit --mem=1g -c 8 -p gpu-troja,gpu-ms --gres=gpu:1 '--constraint=gpuram11G|gpuram16G|gpuram24G|gpuram40G|gpuram48G' /lnet/troja/projects/elitr/2021-12/multi-source-training-repo/sota-online-asr-230330/.qsubmit-6kn39izx.bash
Submitted job ID: 4048509

CORRECT:

qsubmit -q gpu-* -gpus=1 -M 48g -c 8 -mem 20g --logdir=logs-faster "$x"
sbatch -o logs-faster/qsubmit.o%j -J qsubmit --mem=20g -c 8 -p gpu-troja,gpu-ms --gres=gpu:1 --constraint=gpuram48G /lnet/troja/projects/elitr/2021-12/multi-source-training-repo/sota-online-asr-230330/.qsubmit-7ih1adsd.bash
Submitted job ID: 4048511

The difference makes the parameter --gpus=1. It should be guessed by default because of gpu-* queue.

[Gldkslfmsd]

fixed it today