NVIDIA Open-Sources Run:ai GPU Orchestration Platform and Donates KAI Scheduler to CNCF
NVIDIA has open-sourced its Run:ai AI workload and GPU orchestration platform, donating the core KAI Scheduler to the CNCF Sandbox under Apache 2.0 alongside fo
On October 13, 2026, NVIDIA officially announced that it is transitioning its entire Run:ai AI workload and GPU orchestration platform into open-source software within the DSX AI Factory framework.
Image source: blockchain.news / NVIDIA
This initiative opens up NVIDIA's advanced GPU resource scheduling and cluster virtualization technologies—previously a proprietary cornerstone for managing large-scale AI workloads—to the broader cloud-native ecosystem.
Four Modular Components and CUDA 13.4 Hardware-Level GPU Memory Fractioning
The open-sourced Run:ai platform is structured as four independent, modular components that engineering teams can adopt individually or integrate into existing Kubernetes environments:
- KAI Scheduler: The primary scheduling engine managing workload queuing, priority policies, and preemptive job execution across Kubernetes clusters.
- GPU Fractioning Operator: Built on CUDA 13.4 to deliver hardware-level memory isolation, enabling a single physical GPU to be divided into fine-grained virtual slices with strict isolation guarantees.
- KAI Resource Management: Handles topology-aware GPU resource discovery, node pool partitioning, and quota management.
- Workload Map: An operational visualization component providing real-time visibility into active jobs, queue latency, and GPU cluster utilization.
In particular, the GPU Fractioning Operator module integrates directly with CUDA 13.4 memory isolation capabilities, helping operators mitigate compute waste in fine-tuning and inference pipelines by maximizing fractional GPU density.
CNCF Sandbox Donation and Distribution via GitHub and NGC Community Helm Charts
To ensure vendor neutrality and community governance, NVIDIA has licensed the core KAI Scheduler under the Apache 2.0 license and formally donated it as a Cloud Native Computing Foundation (CNCF) Sandbox project.
Infrastructure teams can access and deploy the platform without requiring a paid subscription through two primary distribution channels:
- Complete Source Code: All four components are publicly accessible in the official NVIDIA DSX AI Factory GitHub repository.
- NGC Community Helm Charts: Ready-to-deploy Helm packages are available through the NVIDIA GPU Cloud (NGC) community catalog without requiring a paid license.
This distribution model allows enterprises and research organizations operating self-managed Kubernetes infrastructure to deploy production-grade GPU scheduling without proprietary software lock-in.
Enterprise Tier Separation and Operational Caveats
Alongside the open-source release, NVIDIA clarified the structural boundaries between community distributions and its commercial offerings:
- Commercial Lineup Maintained: NVIDIA AI Enterprise and Mission Control commercial editions remain separate offerings tailored for mission-critical deployments requiring enterprise-grade technical support.
- Support and SLA Limitations: Neither the open-source GitHub code nor the NGC Community Helm distributions include official NVIDIA technical support or SLAs. Organizations requiring 24/7 incident response must maintain commercial enterprise licenses.
- License Acceptance: Utilizing the NGC Community distribution packages requires agreeing to the NVIDIA Software License Agreement (SLA) and specific product terms.
Sources
- Blockchain.news: NVIDIA Releases Run:ai GPU Orchestration Platform as Open Source