Skip to content

Research Computing User Guide

Welcome. This guide shows you how to use the UMD Engineering research computing portal ("OOD" — Open OnDemand) to run your lab's software on GPU workstations without having to manage any Linux hosts yourself.

What this system is

This isn't a campus-wide shared HPC pool. Each participating lab owns its own set of GPU nodes (typically 1–5 per lab). All labs sit behind the same web portal and the same Slurm scheduler, but your jobs only ever run on your lab's nodes, and the only people you queue alongside are your own labmates.

  • A web portal at https://hpc.eng.umd.edu where you sign in with your UMD directory ID.
  • Behind the portal, Slurm assigns your session to one of your lab's GPU compute nodes — never another lab's.
  • The session gives you a full Linux desktop in your browser with GPU acceleration, your lab's software pre-installed, and your lab's storage already mounted.
  • When you're done, the session ends and the node goes back into your lab's pool. Files on lab storage stay put.

Who can use it

Researchers whose PIs have onboarded their lab. If you're not sure, ask your PI; if your PI is reading this, ping the IT team.

Start here

New here? Read getting-started.md — it's a single page covering everything you need for your first session: prerequisites, the three ways to use the system, end-to-end walkthroughs for desktop, Jupyter, and CLI batch jobs, and storage basics.

Three ways to use the system

Mode What it is Best for
Browser desktop OOD portal → Interactive Apps → Lab Desktop (or Lab Desktop Advanced) Graphical apps (MATLAB, ANSYS, FSLeyes), 3D viz, anything you'd double-click
JupyterLab OOD portal → Interactive Apps → JupyterLab Python notebooks, ML training with GPU, prototyping
VS Code (code-server) OOD portal → Interactive Apps → VS Code (code-server) Full VS Code in your browser, running on your lab's GPU node via Slurm. The supported VS Code workflow — don't point desktop VS Code Remote-SSH at the cluster (it lands on the OOD/Slurm controller and degrades the scheduler).
Direct SSH ssh <user>@hpc.eng.umd.edu (submit host) Batch ops only — sbatch, squeue, sacct, editing sbatch scripts

One rule across all three: run compute jobs through Slurm. Inside a desktop or Jupyter session you already are. From SSH, use sbatch for batch work or srun --pty bash for an interactive allocation — never run heavy work as a bare python … outside a Slurm allocation.

Deep-dive guides

Guide Read it when
launching-a-desktop.md You want a GPU desktop, including native VNC client setup
your-lab-storage.md Where your files live; how they follow you
software.md Loading MATLAB, ANSYS, FreeSurfer, Python, etc.
python-and-conda.md Building PyTorch / JAX / conda envs
slurm-cli.md Batch jobs, checking what's available, right-sizing
direct-ssh.md Submit-host SSH for batch ops only (sbatch / squeue / sacct). For VS Code use the OOD code-server tile, not Remote-SSH.
faq.md Common questions and gotchas

Quick-reference shortcuts

  • Portal: https://hpc.eng.umd.edu
  • Maintenance window: third Saturday of every month, 08:00 – 12:00 ET. Nodes reboot; running sessions are terminated. OOD shows a banner in the days leading up. Save work in advance.
  • Your home directory is local to the node your session lands on. Each lab has its own dedicated GPU nodes (typically 1–5). When you launch an OOD session, the profile you pick (e.g. Inspire Turing, Inspire Searle) pins you to one specific node — pick the same profile next time and you'll see the same ~. Pick a different profile and you'll get a different node with its own fresh ~. Either way, keep your real work on your lab share — see your-lab-storage.md.
  • Software: module avail shows what's installed; module load <app> puts it on your PATH.
  • Containers: module load freesurfer, module load fmriprep, module load cpac — they run via Apptainer transparently.

Support

  • Your lab's IT liaison or PI first — for "how does my lab do X," lab AD group additions, and lab-specific workflow questions.
  • Engineering IT (eit-help@umd.edu) for portal, desktop, storage, software requests, and anything broken.