Slurm powers most of the world's top supercomputers with capabilities beyond native Kubernetes: gang scheduling, topology-aware placement, fair-share policies, and job accounting. As AI scales up, these ecosystems can no longer operate in silos — Kubernetes gains battle-tested HPC scheduling, the HPC community gains cloud-native operations. It's time they converge.

We share the journey of bringing Slurm to Kubernetes through Slinky, an open-source operator that runs Slurm natively as Kubernetes workloads. I cover the obstacles — keeping state in sync across systems, handling node failures, enabling high-bandwidth GPU communication — and how we solved them. Whether you run four GPUs or four thousand, the patterns apply.

Attendees will learn why HPC scheduling and Kubernetes are stronger together, how to deploy a Slurm-enabled cluster with open-source Helm charts, and lessons from production. Whether you're a platform engineer, cluster admin, or just curious, this session is for you.