DRA · Dynamic Resource Allocation
The Kubernetes mechanism that replaces counting opaque devices with claiming them by attribute, letting a workload request GPUs by memory, topology, or partition rather than by integer count.
Current numbers
K8s 1.32+Standalone DRA driver path for ComputeDomains needs Kubernetes 1.32+, with the DRA feature gate enabled on 1.32 and 1.33
1 / nodeKubernetes DRA driver allows one ComputeDomain per node, so a second domain cannot use that node's leftover GPUs
GA in 1.34Core Kubernetes DRA graduated to stable in 1.34, released 27 August 2025