From 689cf14202f8bfbf85931c1cae7d4a49121ee725 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Martin=20Lilleeng=20S=C3=A6tra?= Date: Mon, 24 Jan 2022 18:17:41 +0100 Subject: [PATCH 1/2] Update README.md --- README.md | 2 ++ 1 file changed, 2 insertions(+) diff --git a/README.md b/README.md index ce0eb65..f274579 100644 --- a/README.md +++ b/README.md @@ -21,6 +21,8 @@ Example job script: dgx-2-test.job Submit: +module use /cm/shared/ex3-modules/latest/modulefiles # Latest ex3-modules +module load slurm/20.02.7 # To load slurm module sbatch dgx-2-test.job ### PPI 4 x P100 (VPN necessary) From bc39170efb307da56d87ede1b4fd3580e3507ac8 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Martin=20Lilleeng=20S=C3=A6tra?= Date: Mon, 24 Jan 2022 19:56:07 +0100 Subject: [PATCH 2/2] Tweaked settings for DGX-2 run. --- saga-dev.job | 15 ++++++++------- saga-test.job | 11 ++++++----- saga_strong_scaling_benchmark.job | 2 +- 3 files changed, 15 insertions(+), 13 deletions(-) diff --git a/saga-dev.job b/saga-dev.job index 1244048..a0023ff 100644 --- a/saga-dev.job +++ b/saga-dev.job @@ -3,22 +3,23 @@ #SBATCH --job-name=ShallowWaterGPUScalingDev # # Project: -#SBATCH --account=nn9550k +#SBATCH --account=nn9882k # # Wall clock limit: -#SBATCH --time=01:00:00 +#SBATCH --time=00:20:00 +# +# NOTE: See https://documentation.sigma2.no/jobs/projects_accounting.html when adjusting the values below # -# Ask for 1 GPU (max is 2) # Note: The environment variable CUDA_VISIBLE_DEVICES will show which GPU # device(s) to use. It will have values '0', '1' or '0,1' corresponding to # /dev/nvidia0, /dev/nvidia1 or both, respectively. -#SBATCH --partition=accel --gres=gpu:1 +#SBATCH --partition=accel # # Max memory usage per task (core) - increasing this will cost more core hours: -#SBATCH --mem-per-cpu=16G +#SBATCH --mem-per-cpu=3800M # # Number of tasks: -#SBATCH --nodes=1 --ntasks-per-node=1 +#SBATCH --nodes=1 --gpus-per-node=1 --ntasks-per-node=1 # #SBATCH --qos=devel @@ -49,5 +50,5 @@ savefile "$SCRATCH/ShallowWaterGPU/*.json" ## Do some work: cd $SCRATCH/ShallowWaterGPU srun $HOME/.conda/envs/ShallowWaterGPU_HPC/bin/python3 --version -srun $HOME/.conda/envs/ShallowWaterGPU_HPC/bin/python3 mpiTesting.py -nx 8192 -ny 8192 --profile +srun $HOME/.conda/envs/ShallowWaterGPU_HPC/bin/python3 mpiTesting.py -nx 1024 -ny 1024 --profile diff --git a/saga-test.job b/saga-test.job index 0b5e9f5..9522c1e 100644 --- a/saga-test.job +++ b/saga-test.job @@ -3,22 +3,23 @@ #SBATCH --job-name=ShallowWaterGPUStrongScaling # # Project: -#SBATCH --account=nn9550k +#SBATCH --account=nn9882k # # Wall clock limit: #SBATCH --time=24:00:00 # -# Ask for 1 GPU (max is 2) +# NOTE: See https://documentation.sigma2.no/jobs/projects_accounting.html when adjusting the values below +# # Note: The environment variable CUDA_VISIBLE_DEVICES will show which GPU # device(s) to use. It will have values '0', '1' or '0,1' corresponding to # /dev/nvidia0, /dev/nvidia1 or both, respectively. -#SBATCH --partition=accel --gres=gpu:1 +#SBATCH --partition=accel # # Max memory usage per task (core) - increasing this will cost more core hours: -#SBATCH --mem-per-cpu=16G +#SBATCH --mem-per-cpu=3800M # # Number of tasks: -#SBATCH --nodes=1 --ntasks-per-node=1 +#SBATCH --nodes=1 --gpus-per-node=1 --ntasks-per-node=1 ## Set up job environment: (this is done automatically behind the scenes) ## (make sure to comment '#' or remove the following line 'source ...') diff --git a/saga_strong_scaling_benchmark.job b/saga_strong_scaling_benchmark.job index 05320d7..0d729c3 100644 --- a/saga_strong_scaling_benchmark.job +++ b/saga_strong_scaling_benchmark.job @@ -3,7 +3,7 @@ #SBATCH --job-name=ShallowWaterGPUStrongScaling # # Project: -#SBATCH --account=nn9550k +#SBATCH --account=nn9882k # # Wall clock limit: #SBATCH --time=10:00:00