NVIDIA announced at KubeCon Europe 2026 in Amsterdam that it will donate the GPU Dynamic Resource Allocation (DRA) Driver to the CNCF. The DRA driver replaces the legacy integer-based GPU resource model in Kubernetes with a richer ResourceClaims-based model, enabling workloads to request specific GPU topologies, NVLink-connected GPUs, GPU slices, and multi-node memory domains. The post also covers the broader NVIDIA GPU stack for Kubernetes: the GPU Operator (automating driver/runtime lifecycle), the KAI Scheduler (fractional GPU allocation, gang scheduling, hierarchical queuing), and how Canonical Kubernetes serves as a production-ready platform for these AI infrastructure components.
Table of contents
Previewing at KubeCon 2026: Canonical welcomes NVIDIA’s donation of the GPU DRA driver to CNCF.The NVIDIA GPU ecosystem for kubernetesCanonical Kubernetes: a platform for cloud-native AI infrastructure14 Impressions