Skip to content

Resource Management Policy

Isambard-AI

Isambard-AI is one of the world’s first, large-scale, open AI supercomputers. Bristol Centre for Supercomputing (BriCS) aims provide a user experience that meets the needs of AI researchers and workloads, which may be quite different from those of simulation-focused high performance computing (HPC). To achieve this, a set of guiding principles for internal technical decisions regarding compute resource allocation on the platform have been defined:

  • Both interactive and batch workloads are welcome on the Isambard-AI compute nodes.
  • Near-instant shared use of CPUs and GPUs for non-compute-intensive work will be provided where possible.
  • Faster access to smaller amount of compute, e.g. for interactive use or exploratory development, will be prioritised over total platform utilisation.
  • For larger workloads, users may submit any technically-sound resource request that is within the bounds of:
    • Project GPU-hour allocation as set by the awarding body.
    • Per-user, per-project or other queue limits as set by the BriCS team.
  • Administrators reserve the right to manage, suspend or cancel jobs to protect the overall health of the service.

These principles will be reviewed from time to time as the platform and its users evolve.