smyalygames/FiniteVolumeGPU
1
0
Fork 0
mirror of https://github.com/smyalygames/FiniteVolumeGPU.git synced 2025-05-17 22:14:14 +02:00
Code Issues Actions Packages Projects Releases Wiki Activity

Added benchmarks for Saga and updated figures.

Browse Source
This commit is contained in:
Martin Lilleeng Sætra 2022-06-24 17:14:10 +02:00
parent 528b738dda
commit a7a723aca6
4 changed files with 304 additions and 247 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

436
Figures.ipynb
View File

File diff suppressed because one or more lines are too long

26
saga_scaling_benchmark.job
View File

@ -16,7 +16,8 @@
#SBATCH --partition=accel
#
# Max memory usage per task (core) - increasing this will cost more core hours:
#SBATCH --mem-per-cpu=3800M
##SBATCH --mem-per-cpu=3800M
#SBATCH --mem-per-cpu=24G
#
#SBATCH --qos=devel
@ -26,6 +27,8 @@
module restore system # instead of 'module purge' rather set module environment to the system default
module load CUDA/11.4.1
#module load CUDA/11.1.1-GCC-10.2.0
#module load OpenMPI/4.0.5-gcccuda-2020b
# It is also recommended to to list loaded modules, for easier debugging:
module list
@ -40,12 +43,23 @@ cp -r . $SCRATCH/ShallowWaterGPU
## Make sure the results are copied back to the submit directory (see Work Directory below):
# chkfile MyResultFile
# chkfile is replaced by 'savefile' on Saga
savefile "$SCRATCH/ShallowWaterGPU/*.log"
savefile "$SCRATCH/ShallowWaterGPU/*.nc"
savefile "$SCRATCH/ShallowWaterGPU/*.json"
#savefile "$SCRATCH/ShallowWaterGPU/*.log"
#savefile "$SCRATCH/ShallowWaterGPU/*.nc"
#savefile "$SCRATCH/ShallowWaterGPU/*.json"
#savefile "$SCRATCH/ShallowWaterGPU/*.qdrep"
cleanup "rm -rf $SCRATCH/ShallowWaterGPU"
export OMPI_MCA_opal_cuda_support=true
## 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 $NX -ny $NY --profile
srun /cluster/projects/nn9882k/martinls/.conda/envs/ShallowWaterGPU_HPC/bin/python3 --version
srun /cluster/projects/nn9882k/martinls/.conda/envs/ShallowWaterGPU_HPC/bin/python3 mpiTesting.py -nx $NX -ny $NY --profile
cd $HOME/src/ShallowWaterGPU
mkdir -p output_saga/$NOW/$SLURM_JOB_ID
mv $SCRATCH/ShallowWaterGPU/*.log ./output_saga/$NOW/$SLURM_JOB_ID
mv $SCRATCH/ShallowWaterGPU/*.nc ./output_saga/$NOW/$SLURM_JOB_ID
mv $SCRATCH/ShallowWaterGPU/*.json ./output_saga/$NOW
mv $SCRATCH/ShallowWaterGPU/*.qdrep ./output_saga/$NOW

36
saga_strong_scaling_benchmark.sh
View File

@ -1,12 +1,30 @@
#!/bin/bash
# one node: 1-4 GPUs
sbatch --nodes=1 --gpus-per-node=1 --ntasks-per-node=1 --export=ALL,NX=8192,NY=8192 saga_scaling_benchmark.job
sbatch --nodes=1 --gpus-per-node=2 --ntasks-per-node=2 --export=ALL,NX=8192,NY=4096 saga_scaling_benchmark.job
sbatch --nodes=1 --gpus-per-node=3 --ntasks-per-node=3 --export=ALL,NX=8192,NY=2731 saga_scaling_benchmark.job
sbatch --nodes=1 --gpus-per-node=4 --ntasks-per-node=4 --export=ALL,NX=8192,NY=2048 saga_scaling_benchmark.job
TIMESTAMP=$(date "+%Y-%m-%dT%H%M%S")
# 2-4 nodes: 1 GPUs per node
sbatch --nodes=2 --gpus-per-node=1 --ntasks-per-node=1 --export=ALL,NX=8192,NY=4096 saga_scaling_benchmark.job
sbatch --nodes=3 --gpus-per-node=1 --ntasks-per-node=1 --export=ALL,NX=8192,NY=2731 saga_scaling_benchmark.job
sbatch --nodes=4 --gpus-per-node=1 --ntasks-per-node=1 --export=ALL,NX=8192,NY=2048 saga_scaling_benchmark.job
# one node: 1–4 GPUs
sbatch --nodes=1 --gpus-per-node=1 --ntasks-per-node=1 --export=ALL,NX=20480,NY=20480,NOW=$TIMESTAMP saga_scaling_benchmark.job # 1 ranks
sbatch --nodes=1 --gpus-per-node=2 --ntasks-per-node=2 --export=ALL,NX=20480,NY=10240,NOW=$TIMESTAMP saga_scaling_benchmark.job # 2 ranks
sbatch --nodes=1 --gpus-per-node=3 --ntasks-per-node=3 --export=ALL,NX=20480,NY=6826,NOW=$TIMESTAMP saga_scaling_benchmark.job # 3 ranks
sbatch --nodes=1 --gpus-per-node=4 --ntasks-per-node=4 --export=ALL,NX=20480,NY=5120,NOW=$TIMESTAMP saga_scaling_benchmark.job # 4 ranks
# 4 nodes: 1–4 GPUs per node
sbatch --nodes=4 --gpus-per-node=1 --ntasks-per-node=1 --export=ALL,NX=20480,NY=5120,NOW=$TIMESTAMP saga_scaling_benchmark.job # 4 ranks
sbatch --nodes=4 --gpus-per-node=2 --ntasks-per-node=2 --export=ALL,NX=20480,NY=2560,NOW=$TIMESTAMP saga_scaling_benchmark.job # 8 ranks
sbatch --nodes=4 --gpus-per-node=3 --ntasks-per-node=3 --export=ALL,NX=20480,NY=1706,NOW=$TIMESTAMP saga_scaling_benchmark.job # 12 ranks
sbatch --nodes=4 --gpus-per-node=4 --ntasks-per-node=4 --export=ALL,NX=20480,NY=1280,NOW=$TIMESTAMP saga_scaling_benchmark.job # 16 ranks
# 4 nodes: 1–4 GPUs per node
sbatch --nodes=4 --gpus-per-node=1 --ntasks-per-node=1 --export=ALL,NX=40960,NY=10240,NOW=$TIMESTAMP saga_scaling_benchmark.job # 4 ranks
sbatch --nodes=4 --gpus-per-node=2 --ntasks-per-node=2 --export=ALL,NX=40960,NY=5120,NOW=$TIMESTAMP saga_scaling_benchmark.job # 8 ranks
sbatch --nodes=4 --gpus-per-node=3 --ntasks-per-node=3 --export=ALL,NX=40960,NY=3413,NOW=$TIMESTAMP saga_scaling_benchmark.job # 12 ranks
sbatch --nodes=4 --gpus-per-node=4 --ntasks-per-node=4 --export=ALL,NX=40960,NY=2560,NOW=$TIMESTAMP saga_scaling_benchmark.job # 16 ranks
## one node: 1–4 GPUs
#sbatch --nodes=1 --gpus-per-node=4 --ntasks-per-node=4 --export=ALL,NX=24576,NY=6144,NOW=$TIMESTAMP saga_scaling_benchmark.job # 4 ranks
#
## 4 nodes: 1–4 GPUs per node
#sbatch --nodes=4 --gpus-per-node=1 --ntasks-per-node=1 --export=ALL,NX=24576,NY=6144,NOW=$TIMESTAMP saga_scaling_benchmark.job # 4 ranks
#sbatch --nodes=4 --gpus-per-node=2 --ntasks-per-node=2 --export=ALL,NX=24576,NY=3072,NOW=$TIMESTAMP saga_scaling_benchmark.job # 8 ranks
#sbatch --nodes=4 --gpus-per-node=3 --ntasks-per-node=3 --export=ALL,NX=24576,NY=2048,NOW=$TIMESTAMP saga_scaling_benchmark.job # 12 ranks
#sbatch --nodes=4 --gpus-per-node=4 --ntasks-per-node=4 --export=ALL,NX=24576,NY=1536,NOW=$TIMESTAMP saga_scaling_benchmark.job # 16 ranks

25
saga_weak_scaling_benchmark.sh Normal file
View File

@ -0,0 +1,25 @@
#!/bin/bash
TIMESTAMP=$(date "+%Y-%m-%dT%H%M%S")
# one node: 1-4 GPUs
sbatch --nodes=1 --gpus-per-node=1 --ntasks-per-node=1 --export=ALL,NX=20480,NY=20480,NOW=$TIMESTAMP saga_scaling_benchmark.job # 1 ranks
sbatch --nodes=1 --gpus-per-node=2 --ntasks-per-node=2 --export=ALL,NX=20480,NY=20480,NOW=$TIMESTAMP saga_scaling_benchmark.job # 2 ranks
sbatch --nodes=1 --gpus-per-node=3 --ntasks-per-node=3 --export=ALL,NX=20480,NY=20480,NOW=$TIMESTAMP saga_scaling_benchmark.job # 3 ranks
sbatch --nodes=1 --gpus-per-node=4 --ntasks-per-node=4 --export=ALL,NX=20480,NY=20480,NOW=$TIMESTAMP saga_scaling_benchmark.job # 4 ranks
# 2-4 nodes: 1 GPUs per node
sbatch --nodes=2 --gpus-per-node=1 --ntasks-per-node=1 --export=ALL,NX=20480,NY=20480,NOW=$TIMESTAMP saga_scaling_benchmark.job # 2 ranks
sbatch --nodes=3 --gpus-per-node=1 --ntasks-per-node=1 --export=ALL,NX=20480,NY=20480,NOW=$TIMESTAMP saga_scaling_benchmark.job # 3 ranks
sbatch --nodes=4 --gpus-per-node=1 --ntasks-per-node=1 --export=ALL,NX=20480,NY=20480,NOW=$TIMESTAMP saga_scaling_benchmark.job # 4 ranks
## one node: 1-4 GPUs
#sbatch --nodes=1 --gpus-per-node=1 --ntasks-per-node=1 --export=ALL,NX=12288,NY=12288,NOW=$TIMESTAMP saga_scaling_benchmark.job # 1 ranks
#sbatch --nodes=1 --gpus-per-node=2 --ntasks-per-node=2 --export=ALL,NX=12288,NY=12288,NOW=$TIMESTAMP saga_scaling_benchmark.job # 2 ranks
#sbatch --nodes=1 --gpus-per-node=3 --ntasks-per-node=3 --export=ALL,NX=12288,NY=12288,NOW=$TIMESTAMP saga_scaling_benchmark.job # 3 ranks
#sbatch --nodes=1 --gpus-per-node=4 --ntasks-per-node=4 --export=ALL,NX=12288,NY=12288,NOW=$TIMESTAMP saga_scaling_benchmark.job # 4 ranks
## 2-4 nodes: 1 GPUs per node
#sbatch --nodes=2 --gpus-per-node=1 --ntasks-per-node=1 --export=ALL,NX=12288,NY=12288,NOW=$TIMESTAMP saga_scaling_benchmark.job # 2 ranks
#sbatch --nodes=3 --gpus-per-node=1 --ntasks-per-node=1 --export=ALL,NX=12288,NY=12288,NOW=$TIMESTAMP saga_scaling_benchmark.job # 3 ranks
#sbatch --nodes=4 --gpus-per-node=1 --ntasks-per-node=1 --export=ALL,NX=12288,NY=12288,NOW=$TIMESTAMP saga_scaling_benchmark.job # 4 ranks