NVIDIA opens the full Run:ai platform for shared GPU workloads

NVIDIA’s October 13 announcement makes the complete platform for sharing graphics processors (GPUs) open source. Teams can adopt it together or use components separately. The Community Product on NGC includes an installation chart but does not include NVIDIA enterprise support.

The repository explains how work moves through the system. Users submit workloads through a browser interface, command line or API. A central management layer handles resources and policies across connected Kubernetes clusters; services in each cluster receive requests and apply scheduling policies. Status returns to the central layer. Run:ai runs above Kubernetes rather than replacing it. It addresses teams sharing GPUs and enforcing resource quotas; for one team with dedicated GPUs, ordinary Kubernetes scheduling may already suffice.

The local development guide describes an environment that simulates GPUs by default, allowing exploration without physical GPU hardware. Its prerequisites matter: only macOS or Linux on amd64 or arm64 is supported. Docker or Podman must be installed and running; Linux also needs the containerd image store enabled. The host must provide make, git, curl, python3 and openssl. These tools are not installed automatically, and missing ones stop setup. This prerequisite overview concerns the development environment; simulated hardware does not demonstrate production performance.

MarkTechPost also covered the release.