Making Kubernetes AI-Ready for NVIDIA Vera Rubin on Day One
Web page · https://developer.nvidia.com/blog/making-kubernetes-ai-ready-for-nvidia-vera-rubin-on-day-one
Configuring and optimizing Kubernetes for new GPU hardware such as NVIDIA Vera Rubin is complex and error-prone. Drivers, kernels, runtimes, networking,...
First seen by Oriel on .
AI summary
This NVIDIA post presents AICR recipes that configure and validate Kubernetes for NVIDIA Vera Rubin NVL72, with end-to-end GPU Operator, DRA, IMEX, and NCCL checks passing on early hardware.
- AICR captures driver, runtime, networking, scheduling, and topology choices in versioned recipes, then generates deployment bundles and validates cluster state.
- Vera Rubin requires driver R615.23 in testing, ConnectX-9 SuperNICs, BlueField-4 DPUs, DOCA 3.3.0, and topology-specific GPU-to-NIC and NUMA checks.
- The Preview recipe targets RKE2, Ubuntu 26.04, Kubernetes 1.34.1 or newer, a 64k-page arm64 kernel, host-managed drivers, and DRA ComputeDomains.
- Training validation passed deployment and conformance checks, including cross-node NCCL all-reduce over NVLink at 744.71 GB/s bus bandwidth with 16 GiB.
- Validation uses whole-GPU allocation; Vera Rubin MIG is not covered, and RDMA data-path performance over ConnectX-9 remains outside the validation scope.
#kubernetes #gpu operator #dynamic resource allocation #nvidia nvlink
Generated by gpt-5.6-sol. Check the original for details.
Aliases
| Source | External ID | URL |
|---|---|---|
| NVIDIA Developer Blog | https://developer.nvidia.com/blog/?p=124438 | https://developer.nvidia.com/blog/making-kubernetes-ai-ready-for-nvidia-vera-rubin-on-day-one/ Read discussion |