Having GPU quota doesn't guarantee that a training job can run. In this on-demand session, Cohere's John Jordan and CoreWeave's Amy Chen explore how Kueue's Topology Aware Scheduling brings the physical realities of modern GPU clusters into Kubernetes. Learn how NVLink domains, rack boundaries, and cluster fragmentation affect workload placement and performance, along with the configuration choices that matter, including preferred versus required topologies and PodSet slices. The session also covers practical production lessons around sticky scheduling, quota debugging, fragmentation, and node hot swaps.
Speakers: Amy Chen, Senior Software Engineer, CoreWeave; John Jordan, Member of Technical Staff, Cohere
