GPU Computing

Use GPU-accelerated resources for AI, image analysis, molecular modelling, deep learning and scientific workloads designed to run on GPUs.

Use the GPU Computing service when your workflow can benefit from GPU acceleration and requires more GPU memory, runtime or shared infrastructure than a personal workstation can provide.

Use this service when you need to:

train or run AI and deep learning models;
process large image datasets;
run GPU-enabled scientific software;
use CUDA, PyTorch, TensorFlow or other GPU-aware frameworks;
run workloads that need more GPU memory than a personal workstation;
execute GPU jobs close to GPFS research storage;
scale from interactive testing to scheduled GPU jobs through Slurm.

What you can run

Artificial intelligence and deep learning

Training, inference and experimentation with GPU-enabled AI frameworks such as PyTorch or TensorFlow.

Image analysis and visualization

Large-scale image processing, microscopy workflows, medical imaging and graphical scientific applications.

Molecular modelling and scientific simulation

GPU-enabled applications for molecular dynamics, modelling and other accelerated scientific workloads.

Data-intensive GPU workflows

Analyses where GPU acceleration reduces runtime for large datasets or repeated computations.

Interactive GPU development

Testing notebooks, scripts and applications interactively before scaling them as scheduled jobs.

GPU partitions

Standard GPU partition

Purpose: Interactive, graphical and standard GPU-accelerated workloads.

Partition: std-gpu

Hardware summary:

  • NVIDIA L40S GPUs
  • 1 GPU per node
  • 2 TB RAM per node

Use when: Your workload needs one GPU, interactive development, visualization, notebooks or standard GPU acceleration.

High GPU partition

Purpose: Demanding GPU workloads, multi-GPU jobs and larger accelerated workflows.

Partition: high-gpu

Hardware summary:

  • NVIDIA L40S GPUs
  • 4 GPUs per node
  • 1 TB RAM per node

Use when: Your workflow needs multiple GPUs, larger GPU jobs or heavier AI, imaging or scientific workloads.

If you are not sure which GPU partition to use, contact SCC before running large workloads.