NVIDIA open GPU driver donation
NVIDIA donated its open‑source GPU driver for AI production to the CNCF, a move aimed at deeper Kubernetes integration and smoother cloud‑native GPU workflows (x.com). That makes it easier for dev teams to run GPU workloads inside CNCF stacks and could lower integration friction for startups shipping AI services (x.com).
The release carries a formal name: NVIDIA’s codebase is now published as the "NVIDIA DRA Driver for GPUs" (Dynamic Resource Allocation) and was highlighted in NVIDIA’s KubeCon Europe writeup on March 24, 2026 in Amsterdam. (blogs.nvidia.com) The DRA driver exposes two resource types—GPUs and ComputeDomains—and is built to enable live GPU allocation and dynamic reconfiguration, including explicit support for multi-node NVLink topologies. ( ) NVIDIA’s documentation and repo show the driver splits functionality into two DRA kubelet plugins (gpu-kubelet-plugin and compute-domain-kubelet-plugin) and documents compatibility requirements such as Kubernetes v1.34.2 or newer for full feature support. ( ) The company also moved related orchestration pieces: its KAI Scheduler repository (now maintained publicly) has been accepted into the CNCF Sandbox and NVIDIA introduced Grove, a Kubernetes API for coordinating AI/GPU workloads, at the same conference. ( ) NVIDIA announced additional upstream work at KubeCon, including GPU support for Kata Containers (confidential containers) and partner collaboration from vendors cited in event coverage such as Amazon Web Services, Google Cloud, Microsoft, Broadcom, Canonical and SUSE. ( ) The driver repository is available under an open-source license on GitHub, showing over a thousand commits and several hundred stars, and the project README and CI indicate community contribution pathways are open now. (github.com)
