NVIDIA open-sourced Run:ai: what changed for Kubernetes GPU teams

A five-slide summary of NVIDIA's October 13, 2026 Run:ai open-source release: the four components, what most coverage misses (CUDA 13.4 fractions, no enterprise support on the free build), and where to start. Source: https://developer.nvidia.com/blog/nvidia-runai-is-now-open-source-as-part-of-nvidia-dsx-ai-factory-platform/

Transcript

1. NVIDIA open-sourced Run:ai What actually changed for Kubernetes GPU teams Announced October 13, 2026 Source: NVIDIA Technical Blog Clydara 1/5
2. What shipped Four components you can adopt on their own KAI Scheduler: gang scheduling, fair-share, quotas, topology-aware placement (already Apache 2.0, a CNCF Sandbox project) GPU Fractioning Operator: Kubernetes layer for CUDA 13.4 native GPU fractions KAI Resource Management: projects, departments, hierarchical quotas, node pools Workload Map: Go library describing AI workload CRDs once for any tool Clydara 2/5
3. What most coverage misses Check these before you plan around it Proprietary fractioning code was retired, not opened: your CUDA driver decides The free NGC Community build has no NVIDIA enterprise support KAI Scheduler can run alongside an existing stack Clydara 3/5
4. The real change Projects, quotas and node pools were part of what customers paid for They are now readable, forkable code Start small: fair-share between two teams may only need KAI Scheduler Clydara 4/5
5. Read more NVIDIA Technical Blog, Oct 13, 2026: developer.nvidia.com/blog/nvidia-runai-is-now-open-source-as-part-of-nvidia-dsx-ai-factory-platform Full write-up: clydara.wordpress.com Clydara · clydralab.com Clydara 5/5