Nvidia opens sources Run:ai for kubernetes GPU management

NVIDIA has open-sourced its Run:ai platform under the Apache 2.0 license, making it available on GitHub. The platform, designed for Kubernetes GPU orchestration, enables developers to manage GPU scheduling, quotas, and workloads. It is now integrated into the NVIDIA DSX AI Factory platform, allowing teams to inspect, fork, and self-host the stack without a paid subscription. The release includes four components, with a local development guide that builds 50 Go services inside containers and creates a 5-node Kubernetes cluster. The platform supports real NVIDIA GPUs with the --gpu-operator flag, requiring Linux, Docker, and the NVIDIA Container Toolkit. It also includes adjacent projects like Model Streamer and Grove, which enhance GPU memory and multicomponent inference workflows, respectively. Microsoft’s benchmark showed a 3x to 6x faster loading speed using NVIDIA H100 VMs. The release is supported by eight DSX ecosystem partners, including Crusoe and Mirantis, and is available through NGC Community Product Helm charts and commercial editions like NVIDIA AI Enterprise.