Shallow water benchmarking experiment based of https://github.com/climate-machine/Canary.jl/commit/a58ac057a7224f45bcc450ee533937f7bd06fe17
julia --project=. -e "using Pkg; Pkg.instantiate(); Pkg.API.precompile()"
mpirun -n 4 --oversubscribe julia --project=. original/shallow_water.jl
Run ./setup.sh
everytime that you updated a package, on the login node.
run.sc.sh
is setup to start 32 mpi ranks per default,
using up two full nodes. It is important that you run the setup before starting
the job and after the Manifest.toml
got updated since the compute nodes don't
have access to the internet.
sbatch run.sc.sh
run.sc.gpu.sh
is setup to use one node with 4 GPUs, in order to pull in the
GPU functionality we are using a stacked environment.
sbatch run.sc.gpu.sh