Slurm Over Kubernetes for Local GPU Cluster Scheduling

Ubunta · x · 2026-08-17

The author shares experience managing local GPU scheduling for clinical notes. Facing resource contention among teams, Slurm was chosen over Ray, K8s, and Celery. Slurm handles resource allocation efficiently, and its sacct command automatically maintains detailed job history, eliminating the need for a custom tracking system.

Original post →

More from Infra

Infra channel →