Nvidia upgraded Kubernetes provisions for its Vera Rubin platform.

The newly-released Nvidia AI Cluster Runtime (AICR) aims to capture tuned configurations as reusable presets, while generating deployment bundles and validating running clusters to streamline Kubernetes readiness for new GPU hardware.

Nvidia positioned the tool as doing away with the errors and complexity of configuring and optimizing Kubernetes for new chip fleets, making cloud environments ready for Nvidia Vera Rubin NVL72 compute trays upon launch.

AICR was described as the third generation of NVIDIA MGX rack-scale architecture, as originally designed for GB200 NVL72 racks. According to Nvidia, the Vera Rubin NVL72 recipe family encodes validated presets for the host-managed R615.23 driver branch; dynamic resource allocation (DRA) driver v0.5.0 with ComputeDomains; Ubuntu 26.04 operating system with a 64k-page kernel; and platform-specific settings such as node feature discovery (NFD) topology updater memory limits.

Nvidia’s internal testing achieved 744.71 Gb/s bus bandwidth at 16 gibibytes (GiB) over the NVLink fabric, as using GPU Operator v26.7.0, DRA driver v0.5.0; and a cross-node NVIDIA collective communications library (NCCL) performance gate.

AICR depends on a four-step workflow, comprised of recipe generation; deployment bundle rendering; and ComputeDomain testing.

“Standing up a working Vera Rubin cluster means choosing driver mode, container-runtime wiring, component versions, and a handful of platform-specific settings. AICR captures those choices as a versioned recipe, so the configuration on real Vera Rubin hardware is explicit, reproducible, and verifiable across clusters,” Nvidia wrote.

The Vera Rubin recipes can be found in Nvidia’s open-source AICR repository. The chip giant added that the Vera Rubin recipe family was an “early-adopter path that does not yet carry the production support and life cycle qualification of a supported recipe while the platform itself is pre-release.”