Using CPU/GPU resources

The Digital Research Alliance of Canada (formerly Compute Canada) provides Canadian researchers (and their students) with free access to high-performance computing infrastructure. All job submissions go through SLURM, a scheduler that determines when and where your job will run.

→ Access

• If you are a researcher, you need to become a member:
  • https://docs.alliancecan.ca/wiki/Apply_for_a_CCDB_account/en
• If you work with a researcher, you need to become a member and be approved by that researcher (the sponsor).

→ Types of resources

The first step is to determine your needs:

• CPU: Your code is sequential or slightly parallel, uses data processing tools (pandas, R, etc.), runs bioinformatics tools, or lacks GPU-optimized libraries.
• GPU: You train deep learning models (PyTorch, TensorFlow, JAX), run CUDA-accelerated simulations, or perform large-scale matrix operations.

→ Launch the computing job

• The Alliance documentation is the best place to help you:
  https://docs.alliancecan.ca/wiki/Running_jobs/en
• Refer directly to the SLURM documentation if needed:
  https://slurm.schedmd.com/documentation.html

→ Specifics

• You must request exactly the resources you need. Requesting too many lowers your priority (the chance of getting resources in the future).
• Never use the login node (front-end node) to run your jobs. Also, do not use VS Code on the front-end node. Use a compute node instead.
• Some compute nodes do not have internet access.