Dear @Yann.Sagon , you are right, I had two RTX 3090.
Yet is there a way to call a specific GPU ?
And were can I get a glance of GPU utilisation, to adapt my sbatch call ?
The reason:
I actually want to launch job where my interest is the total Vram, that I can share between N GPU. I am trying to get the most disponibility on baobab, to be able to launch multiple jobs, for example ~100 jobs with lets say 30G of Vram.
3x10G would work, but if the RTX 3080 are not available, then I jobs use 3x25, or 3x48 (on node 48), which is a pity and way to much for my job (I would rather use 1x48 for the jobs, and thus get more jobs done for same dispo of Gpus).
Thus my desire to call a specific GPU: In my example, per job, I would call 3x10 RTX 3080 if those are available, or 2x25G RTX A5500 or RTX 3090 if these are the GPU available, or 1x48 RTX A6000 if those are available.
But
#SBATCH --constraint=COMPUTE_MODEL_RTX_3080_10G
Does not work, although you seemed to propose this here: