Interactive GPU on ARCx

Notes on how to set up an interactive GPU session for computation on the Leeds System.

Photo by Pascal Brändle on Unsplash

Start a session

Arc4

qrsh -l h_rt=2:0:0,coproc_v100=1,h_vmem=8G -pty y bash

A single V100 card, 10 CPU cores and 48GB system memory (one quarter of the available resource on a V100 node)

Arc3

qrsh -l h_rt=2:0:0,coproc_k80=1 -pty y bash

A single K80 card, 12 CPU cores and 64GB system memory (half the available resource on a K80 node)

or

qrsh -l h_rt=2:0:0,coproc_p_100=1 -pty y bash

A single P100 card, 6 CPU cores and 64GB system memory (one quarter of the available resource on a P100 node)

Check the Card you Have Been Allocated

module load cuda
nvidia-smi -L

Install libraries you want to use

module load cuda ## add cuda env
mkdir /nobackup/$USER ; cd $_;
conda create --prefix ./gpuenv python=3.8 --yes; 
conda activate /nobackup/$USER/gpuenv;

#add common modules