From b37390bb5a2794c353714175bfb7dba0397753ec Mon Sep 17 00:00:00 2001 From: Peter Boyle Date: Mon, 27 Apr 2026 14:40:11 -0700 Subject: [PATCH] 4 node usqcd run --- systems/Perlmutter/usqcd4.slurm | 39 +++++++++++++++++++++++++++++++++ 1 file changed, 39 insertions(+) create mode 100644 systems/Perlmutter/usqcd4.slurm diff --git a/systems/Perlmutter/usqcd4.slurm b/systems/Perlmutter/usqcd4.slurm new file mode 100644 index 00000000..4a4917cf --- /dev/null +++ b/systems/Perlmutter/usqcd4.slurm @@ -0,0 +1,39 @@ +#!/bin/bash +#SBATCH -A m5294_g +#m3886_g +#SBATCH -C gpu +#SBATCH -q debug +#SBATCH -t 0:30:00 +#SBATCH -c 32 +#SBATCH -N 4 +#SBATCH -n 16 +#SBATCH --ntasks-per-node=4 +#SBATCH --gpus-per-task=1 +#SBATCH --exclusive +#SBATCH --gpu-bind=none + +export SLURM_CPU_BIND="cores" +export MPICH_GPU_SUPPORT_ENABLED=1 +export MPICH_RDMA_ENABLED_CUDA=1 +export MPICH_GPU_IPC_ENABLED=1 +export MPICH_GPU_EAGER_REGISTER_HOST_MEM=0 +export MPICH_GPU_NO_ASYNC_MEMCPY=0 +#export MPICH_SMP_SINGLE_COPY_MODE=CMA + +cat << EOF > select_gpu +#!/bin/bash +export GPU_MAP=(0 1 2 3) +export NUMA_MAP=( 0 1 2 3 ) +export GPU=\$SLURM_LOCALID +export NUMA=\$SLURM_LOCALID +export CUDA_VISIBLE_DEVICES=\$GPU +exec numactl -m \$NUMA -N \$NUMA \$* +EOF + +chmod +x ./select_gpu + +OPT="--comms-sequential --shm-mpi 0" +VOL=64.64.32.32 +srun ./select_gpu ./benchmarks/Benchmark_usqcd --mpi 2.2.2.2 --grid $VOL --device-mem 16000 --accelerator-threads 8 --shm 2048 $OPT > usqcd.log + +