Slurm GPU Resources (Kebnekaise)

Slurm GPU Resources (Kebnekaise)

To request GPU resources one has to include a GRES in the submit file. The general format is:

#SBATCH --gres=gpu:k80:x

where x = 1, 2, or 4.

The gpu enabled nodes contain either two or four K80 cards, each K80 card contains two gpu engines.

On the dual K80 card enabled nodes, one can request a single K80 card (x = 1) or both (x = 2). When requesting a K80 card a whole CPU socket (14 cores) is also dedicated on the same node. The K80 card is connected to the PCI-express bus of the allocated CPU socket.

When requesting four K80 cards, the whole node is allocated (exclusive use).

One can also activate Nvidia Multi Process Service (MPS), if so required, by using:

#SBATCH --gres=gpu:k80:x,mps

 

Updated: 2017-04-20, 16:56