NVIDIA open-sources the full Run:ai platform as part of its DSX AI Factory

NVIDIA

NVIDIA has released the full Run:ai GPU-orchestration platform as open source, in a repository under its DSX AI Factory GitHub organisation. The platform allocates GPUs and manages AI workloads on Kubernetes across data processing, training and inference, and it is also offered on NVIDIA NGC as a Community Product with a Helm chart.

• KAI Scheduler is Apache 2.0 licensed and has been contributed to the CNCF as a Sandbox project.
• Model Streamer reads model weights concurrently to cut cold-start loading; reported load times are 3.5 seconds on WEKA and 1.7 seconds on VAST Data.
• Eight DSX ecosystem partners, including Crusoe, Nebius and vCluster, are integrating the platform.
• The NGC Community Product needs acceptance of NVIDIA's software licence terms and comes without NVIDIA enterprise support.
• Workload Map is deliberately narrow: it does not schedule, queue, autoscale or span clusters, and Grove's Kueue support is still in progress.