RC Services

UVA Research Computing (RC) provides shared HPC and research storage for compute- and data-intensive work. For most data science projects you will use the Rivanna and Afton clusters (scheduled with Slurm) with attached storage options.

Overview: Compute and Storage

UVA Research Computing HPC systems

Compute

Rivanna and Afton together offer hundreds of CPU nodes (tens of thousands of cores), large per-node memory (roughly hundreds of GB up to ~1.5 TB), and GPU nodes for ML/DL workloads. Jobs run through Slurm; interactive work (Jupyter, RStudio, VS Code, Desktop) typically uses the interactive partition via the Open OnDemand web portal.

What you needWhere to start
CPU / memory-heavy analysis or trainingBatch or interactive jobs on Rivanna/Afton
GPU workgpu or interactive partitions; see GPUs on UVA HPC
Short, interactive sessions (code development, debugging)Open OnDemand apps (interactive queue; time and resource limits apply)

Access is allocation-based. Compute time is measured in service units (SUs) (roughly core-hours; GPUs cost more). Common allocation types:

  • Standard: free, renews about yearly, normal queue priority
  • Purchased: paid SUs, higher priority, do not expire
  • Instructional: free, limited SUs for teaching only, normal queue priority; typically expire about two weeks after the class or training ends
  • Dedicated computing: paid, provides exclusive access to nodes of a specific hardware type.

Only faculty can request an allocation. They can add students to the allocation as needed.

Links:

Storage

Several storage options are available on Rivanna and Afton.

LocationQuota (typical)ProvisionedBest forCaveats
/home200 GB, freeAutomatic with your HPC accountScripts, notebooks, light interactive workPersonal only; not ideal for large Slurm I/O
/scratch10 TB, freeAutomatic with your HPC accountActive job inputs/outputs (fast parallel FS)Personal; no backups; files unused ~90 days are deleted
/projectLeased (1 TB+)Optional; purchased by faculty PIShared group data and running HPC jobsResearch Project storage; snapshots; paid
/standardLeased (1 TB+); PIs may get up to 10 TB freeOptional; requested by faculty PILonger-term shared resultsResearch Standard; slower; don’t run jobs here

None of the RC storage offerings provide backups. You should transfer important data to other storage locations on a regular basis.

Only PIs (faculty) can lease or expand group storage via the storage request form. Students cannot place storage change requests.

Full comparison: Storage options

Practical pattern

  1. Keep code and small configs in /home (or a shared /project path).
  2. Stage large data and write job output to /scratch (or /project if your group has it).
  3. Copy important data to /project, /standard, or other storage systems before automatic scratch cleanup deletes them.

Globus is the recommended way to move large datasets between your laptop, cloud or lab storage, and HPC-mounted paths (/home, /scratch, /project, /standard). Use the managed collection UVA Standard Security Storage in the Globus File Manager; for a local machine, install Globus Connect Personal first.

Alternatively, you can use familiar command-line tools such as scp, rsync, rclone, or the AWS CLI.

Working with controlled-access, regulated, or otherwise highly sensitive data

Controlled-access, regulated, or otherwise highly sensitive data must not be processed on Afton/Rivanna. Use RC’s high-security offerings instead (Ivy virtual machines and Rio HPC). When in doubt, ask the RC team which option fits your use case.

Other RC services

Beyond Rivanna/Afton HPC and storage, Research Computing also offers consulting and platform services that many data science projects use alongside the cluster. Browse the full catalog on RC Services. Highlights include:

Don’t know where to start?

Reach out to the RC team for technical support and consultations. See the Getting help page for details.