Nvidia publishes preview Kubernetes configurations for Vera Rubin

Nvidia published preview configurations for running Kubernetes on Vera Rubin NVL72 hardware in its open-source AI Cluster Runtime v1.0.0. The company says training and inference configurations passed deployment and conformance checks on early hardware, with the setup reproduced on a second cluster. The configurations pin component versions and encode driver, networking and scheduling settings that teams can generate, deploy and validate. A training test measured 744.71 GB/s bus bandwidth for a cross-node collective operation over NVLink. The recipes lack production support and lifecycle qualification while the hardware remains pre-release. Inference serving performance testing is deferred because published runtime images do not yet contain the required compiled Rubin kernels.