Pods, Partitions, and Pandemonium: Why Your Job Is Still Pending

A scheduler exists to get out of your way, not stand in it, but without one, it's nearly impossible to keep a scarce GPU resource fully utilized. If your budget isn't unlimited, this session walks through the scheduling technologies that decide whether your job runs now or waits: the differences between Kubernetes, the popular schedulers built on top of it, and Slurm, still the workhorse for many AI teams. You'll leave knowing how to choose the right approach for the jobs and GPUs you're actually managing, including how Mistral uses Slurm to run multiple clusters and user groups without a scheduler team standing between them and their next job.