Skip to main content

Source repo: sdsc-summer-institute-2023 | Branch: main | Last synced: 2026-04-24 10:27:17.425 UTC

High Throughput Computing

Preemptible batch jobs

Preemptible batch jobs behave the same as regular batch jobs, but they may be cancelled (or terminated) by the scheduler at anytime in order to reclaim the compute resources they were provided and redistrbute those resources to run a higher priority job. However, if your workloads are fault-tolerant and can withstand such interruptions, then running your jobs in a preemptible queue or partition can reduce your total compute costs over the lifetime of a project.

Expanse has two (non-refundabled) preemptible partitions that provide you with a 20% service unit (SU) discount.

  • preempt
  • gpu-preempt

https://www.sdsc.edu/support/user_guides/expanse.html

Next - Distributed high-throughput computing