Skip to content

Working on OSC

What this section covers

How to actually run work on OSC: submitting SLURM jobs, managing software environments (modules + venvs), orchestrating multi-step pipelines, and the two lab-specific tooling pages for CARLA and MATLAB/Simulink users.

Prerequisites: you've finished OSC Basics — account is live, SSH works, you can move files. ML-specific stack (PyTorch, PyG, RAPIDS) lives in ML Workflows.

Core topics

  • Job Submission

    SLURM deep-dive: sbatch scripts, partitions, GPU requests, job arrays, debugging failed jobs. The reference page you'll come back to constantly.

    Submit jobs

  • Environment Management

    OSC's module system plus project-level venvs with uv. Cached wheels, $TMPDIR tricks, avoiding home-directory bloat.

    Manage environments

  • Pipeline Orchestration

    Multi-stage experiments with Ray on SLURM — dependency tracking, resource allocation, failure recovery. For when a single sbatch isn't enough.

    Orchestrate pipelines

Specialized tooling

  • CARLA Simulator

    Install, run, and develop against the CARLA autonomous driving simulator on OSC. Headless GPU rendering, scenario scripting, data collection.

    Use CARLA

  • MATLAB & Simulink

    Interactive MATLAB sessions, batch jobs, Simulink models. License pool notes, parallel toolbox on multi-node jobs.

    Run MATLAB

After this section

If your work is ML-heavy, head to ML Workflows for PyTorch, PyG, experiment tracking, and HPC-specific training gotchas.