NVIDIA Releases AICR v1.0 for Stable, Verifiable GPU Cluster Configuration

NVIDIA has released AICR v1.0 (AI Cluster Runtime), a version-locked and validated recipe system for configuring GPU-accelerated Kubernetes clusters. The release addresses the complexity of managing version compatibility across dozens of components, including host kernels, GPU drivers, container runtimes, networking, and storage, which often leads to delayed error tracking and frequent conflicts after deployment.

Core Capabilities and Compatibility

The v1.0 release establishes a stable compatibility contract across the CLI, REST API, Go SDK, bundle layout, and artifact schemas. This ensures that public interfaces remain stable, with breaking changes requiring a major release. The system provides four key functions:

• Snapshot: Records the observed state of the cluster.
• Recipe: Describes the desired configuration and validation steps.
• Bundle: Renders deployment artifacts for tools like Helm, Argo CD, Flux, and Helmfile.
• Validation: Compares the recipe against the running cluster to generate signed evidence.

Supported Infrastructure and Ecosystem

AICR v1.0 supports a broad range of infrastructure, including 10 GPU accelerators (Rubin, Blackwell, Hopper, Ampere, Ada) and 11 Kubernetes services. It is compatible with operating systems such as Ubuntu, COS, Oracle Linux, and Talos (via mixin). Workload support includes Kubeflow and Slurm for training, and Dynamo and NIM for inference.

The runtime integrates with the broader ecosystem through a Pulumi Labs infrastructure-as-code provider and Mirantis k0rdent for multi-cluster management. A public validation dashboard at validation.aicr.run allows users to search recipes by service, GPU, OS, and workload intent. The project has attracted over 100 contributors, with approximately half coming from outside NVIDIA.