Connect to vpn.ruc.dk, then ssh into one of the following targets:
- dirac - main entry point:
ssh <username>@dirac.ruc.dk - i42 & i43 - internet access for file sync (repo cloning, rsync):
ssh -J <username>@dirac.ruc.dk <username>@<i42/i43> - bead50 - RTX 2080 Ti:
ssh -J <username>@dirac.ruc.dk <username>@bead50 - bead67 - RTX 4090:
ssh -J <username>@dirac.ruc.dk <username>@bead67
Alternatively, add this snippet to your ssh config:
Host dirac
HostName dirac.ruc.dk
User <username>
Host i42 i43 bead50 bead67
User <username>
ProxyJump diracand run ssh <target-hostname>.
Set up preferred system sizes in main.py.
While on bead67, run main.py to start the benchmark and save the measurements to csv:
# Example: run gamdpy with autotuned compute plans and save data with the prefix "default"
python3 main.py -p default gamdpy -aYou can use the following arguments:
usage: main.py [-h] [-p [PREFIX]] [-d] {gamdpy,lammps} ...
positional arguments:
{gamdpy,lammps} backend
options:
-h, --help show this help message and exit
-p, --prefix [PREFIX]
identifying prefix for this run (can overwrite data if not unique)
-d, --debug run benchmark with small system sizes (~30-60 seconds per run)
# for gamdpy
usage: main.py gamdpy [-h] [-a]
options:
-h, --help show this help message and exit
-a, --autotune use autotuned compute plans
# for lammps
usage: main.py lammps [-h] [-g]
options:
-h, --help show this help message and exit
-g, --gpu enable GPU-acceleration
Alternatively, edit and run ./batch_run.sh for multiple runs back-to-back.
Data is written to disk after each measurement, so you can check the progress with tail -f data/<filename>.csv.
Use rsync from your local machine to copy files to/from the server:
rsync <source-path> <destination-path>
where a remote path follows the syntax: <username>@dirac.ruc.dk:/absolute/path, or if you've set up your ssh config: dirac:/absolute/path.