10/17/2026
A GPU outside its NVLink island or a TPU on the wrong torus coordinate may be unusable for frontier training.
Join Decheng Dai of Anthropic at KubeCon + CloudNativeCon North America where he will share what worked, what broke and what had to change while building a topology-aware, gang-scheduling Kubernetes scheduler for enormous AI clusters.
Join us November 9-12 in Salt Lake City.
Explore the Session: https://bit.ly/3UARkW9
Register: https://bit.ly/3BDI5XL