Nvidia has released the entire Run:ai platform, which manages AI workloads and GPU resources, as open source.

The company said Run:ai can allocate GPU resources on demand in a Kubernetes environment and support managing the full AI workflow from data processing to training and inference.

With the release, DSX ecosystem partners and AI cloud companies can increase GPU utilisation and operate shared AI infrastructure more efficiently, the company said.

The release covers the full platform and 4 core components that can be used separately. The KAI Scheduler determines job placement so multiple teams can fairly share cluster resources. The GPU Fraction Operator lets multiple jobs share a single GPU. KAI Resource Management sets and guarantees GPU quotas by department and project. Workload Map defines various types of AI workloads running on Kubernetes in a consistent way.

The source code is available from the GitHub DSX AI Factory organisation. It is also available on Nvidia NGC as a community product. The community product can be used without a paid subscription, but does not include Nvidia enterprise support.

Partners also provide Run:ai through their services. Crusoe, Nebius, Mirantis, LypAI, Spectro Cloud, Saturn Cloud, OpenNebula Systems and vCluster have integrated Run:ai or offer it as a managed service. Nebius offers a feature that deploys Run:ai with a single click in its AI cloud.

Nvidia also introduced related open-source projects, Model Streamer and Grove. Model Streamer supports loading model weights quickly into GPU memory to reduce inference start times. Grove manages inference workloads that require multiple components to scale together through a single API.