Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Update perlmutter machine files #295

Merged
merged 1 commit into from
Mar 7, 2024
Merged

Update perlmutter machine files #295

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
17 changes: 14 additions & 3 deletions machines/configure.perlmutter.gpu.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,8 +1,19 @@
#.MF 2024/03/07: At the time we got this to work I had the following modules loaded
#. 1) craype-x86-milan 7) cpe/23.12 13) craype-accel-nvidia80
#. 2) libfabric/1.15.2.0 8) gpu/1.0 14) cray-mpich/8.1.28 (mpi)
#. 3) craype-network-ofi 9) craype/2.7.30 (c) 15) cudatoolkit/12.0 (g)
#. 4) xpmem/2.6.2-2.5_2.38__gd067c3f.shasta 10) cray-dsmml/0.2.2 16) nccl/2.18.3-cu12
#. 5) gcc-native/12.3 11) cray-libsci/23.12.5 (math)
#. 6) perftools-base/23.12.0 12) PrgEnv-gnu/8.5.0 (cpe)
#.Most of these are loaded by default, so we just load some extra/key ones here.
module load PrgEnv-gnu/8.5.0
module load craype-accel-nvidia80
module load cray-mpich/8.1.28
module load cudatoolkit/12.0
module load openmpi/5.0.0rc12
module load nccl/2.18.3-cu12
module unload darshan

: "${PREFIX:=$HOME/gkylsoft}"
./configure CC=nvcc ARCH_FLAGS="-march=native" CUDA_ARCH=80 --prefix=$PREFIX --lapack-inc=$PREFIX/OpenBLAS/include --lapack-lib=$PREFIX/OpenBLAS/lib/libopenblas.a --superlu-inc=$PREFIX/superlu/include --superlu-lib=$PREFIX/superlu/lib/libsuperlu.a --cudamath-libdir=/opt/nvidia/hpc_sdk/Linux_x86_64/23.1/math_libs/12.0/lib64 --use-mpi=yes --mpi-inc=$OPENMPI_ROOT/include --mpi-lib=$OPENMPI_ROOT/lib --use-nccl=yes --nccl-inc=$NCCL_DIR/include --nccl-lib=$NCCL_DIR/lib;

./configure CC=nvcc ARCH_FLAGS="-march=native" CUDA_ARCH=80 --prefix=$PREFIX --lapack-inc=$PREFIX/OpenBLAS/include --lapack-lib=$PREFIX/OpenBLAS/lib/libopenblas.a --superlu-inc=$PREFIX/superlu/include --superlu-lib=$PREFIX/superlu/lib/libsuperlu.a --cudamath-libdir=/opt/nvidia/hpc_sdk/Linux_x86_64/23.1/math_libs/12.0/lib64 --use-mpi=yes --mpi-inc=$CRAY_MPICH_DIR/include --mpi-lib=$CRAY_MPICH_DIR/lib --use-nccl=yes --nccl-inc=$NCCL_DIR/include --nccl-lib=$NCCL_DIR/lib;


51 changes: 51 additions & 0 deletions machines/jobScript.perlmutter-gpu
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,51 @@
#!/bin/bash -l

#.Declare a name for this job, preferably with 16 or fewer characters.
#SBATCH -J <Job Name>
#SBATCH -A <Account Number>

#.Request the queue (enter the possible names, if omitted, default is the default)
#.this job is going to use the default
#SBATCH -q regular

#.Number of nodes to request (Perlmutter has 64 cores and 4 GPUs per node)
#SBATCH -N 2
#SBATCH --ntasks 8

#.Specify GPU needs:
#SBATCH --constraint gpu
#SBATCH --gpus 8

#.Request wall time
#SBATCH -t 00:30:00

#.Mail is sent to you when the job starts and when it terminates or aborts.
#SBATCH --mail-user=<your email>
#SBATCH --mail-type=END,FAIL,REQUEUE

#.Load modules (this must match those in the machines/configure script).
module load PrgEnv-gnu/8.5.0
module load craype-accel-nvidia80
module load cray-mpich/8.1.28
module load cudatoolkit/12.0
module load nccl/2.18.3-cu12

#.On Perlmutter some jobs get warnings about DVS_MAXNODES (used in file stripping).
#.We set it to 24 for now, but really this depends on the amount/size of I/O being performed.
#.See online NERSC docs and the intro_mpi man page.
export DVS_MAXNODES=24_
export MPICH_MPIIO_DVS_MAXNODES=24

#.Run the rt_gk_sheath_2x2v_p1 executable using 1 GPU along x (-c 1) and 8
#.GPUs along the field line (-d 8). See './rt_gk_sheath_2x2v_p1 -h' for
#.more details/options on decomposition. It also assumes the executable is
#.in the present directory. If it isn't, change `./` to point to the
#.directory containing the executable.

echo "srun -u -n 8 --gpus 8 ./rt_gk_sheath_2x2v_p1 -g -M -c 1 -d 8"
srun -u -n 8 --gpus 8 ./rt_gk_sheath_2x2v_p1 -g -M -c 1 -d 8





17 changes: 14 additions & 3 deletions machines/mkdeps.perlmutter.gpu.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,7 +1,18 @@
#.MF 2024/03/07: At the time we got this to work I had the following modules loaded
#. 1) craype-x86-milan 7) cpe/23.12 13) craype-accel-nvidia80
#. 2) libfabric/1.15.2.0 8) gpu/1.0 14) cray-mpich/8.1.28 (mpi)
#. 3) craype-network-ofi 9) craype/2.7.30 (c) 15) cudatoolkit/12.0 (g)
#. 4) xpmem/2.6.2-2.5_2.38__gd067c3f.shasta 10) cray-dsmml/0.2.2 16) nccl/2.18.3-cu12
#. 5) gcc-native/12.3 11) cray-libsci/23.12.5 (math)
#. 6) perftools-base/23.12.0 12) PrgEnv-gnu/8.5.0 (cpe)
#.Most of these are loaded by default, so we just load some extra/key ones here.
module load PrgEnv-gnu/8.5.0
module load craype-accel-nvidia80
module load cray-mpich/8.1.28
module load cudatoolkit/12.0
module load openmpi/5.0.0rc12
module load nccl/2.18.3-cu12
module unload darshan
cd install-deps

: "${PREFIX:=$HOME/gkylsoft}"

cd install-deps
./mkdeps.sh --build-openblas=yes --build-superlu=yes --prefix=$PREFIX --build-openmpi=no MPICC=mpicc MPICXX=mpicxx
Loading