Currently as far as I can tell (I’d be happy to be wrong :smile: ) it is not possible with Baobab to request a (dependent) job with > 8GPUs for an equivalent number of tasks (as this is the max limit of one node). (Related discussion: Interconnect between GPU nodes ) This is the recommended way prop…

How to request > 8 GPU tasks?

Yann.Sagon January 19, 2021, 5:04pm 2

Interconnect between GPU nodes HPC Technical

Update of my post: it is now possible to use --gpus-per-task on Yggdrasil and Baobab. https://hpc-community.unige.ch/t/baobab-scheduled-maintenance-26-27-august-2020/1002/4

Baobab scheduled maintenance: 26-27 August 2020 HPC Announce

Dear all, The maintenance is now over ! And you can use Baobab again. What is new on Baobab and what kept us busy during this maintenance : Slurm 20 : update from Slurm 19.05.7 to Slurm 20.02.4 Slurm Trackable resources (TRES) support. This allows a more flexible use of GPUs, enabling new options for your jobs such are --gpus-per-node, --gpus-per-task, etc. https://slurm.schedmd.com/tres.html https://slurm.schedmd.com/SLUG19/GPU_Scheduling_and_Cons_Tres.pdf Slurm daemon and database now e…

show post in topic

Home
Categories
Guidelines
Terms of Service
Privacy Policy

Powered by Discourse, best viewed with JavaScript enabled