Sleep Revolution computing cluster
The first step for users is to make sure their working environment fits their needs, might have to be reconfigure.
User can install different version of Python and all necessary modules for their work.
We recommend using Python virtual environment, but of course Python modules can be installed without it
Creating your virtual environment
- Python3 has a module venv which can be used to create your environment (python3 -m venv myProject)
- Another way is to install virtualenv with Pip (pip install virtualenv)
- Miniconda or Ancaonda
User can also download and install :
An example setup for Torch
- python3 -m venv myproject
- source myproject/bin/activate
- pip install torch wheel
Please make sure your script can communicate with the GPU, do a test-script before running your project
Useful commands
| cmd | Descr |
|---|---|
| squeue | Show all jobs on the queue |
| sinfo | Information about the cluster |
| sbatch <job.sh> | Submit job |
| sacct | My jobs |
| srun | Run job interactively |
| scancel <id> | kill job with <id> |
| nvidia-smi | info about GPU |
The template script file looks like this
#!/bin/bash
#SBATCH --account=staff
#SBATCH --job-name=MyJob
#SBATCH --gpus-per-node=1
#SBATCH --mem-per-cpu=2G
#SBATCH --partition=main
#SBATCH --output=myBatch.log
Installed software and drivers
* NVIDIA V100 GPU drivers
* CuDA toolkit [version 12.2]
* Python 3.10.12
* pip 22.0.2