Kubernetes 1.34 brings serious heat for anyone juggling GPUs or accelerators. Meet Dynamic Resource Allocation (DRA)—a new way to schedule hardware like you mean it.

DRA adds ResourceClaims, DeviceClasses, and ResourceSlices, slicing device management away from pod specs. It replaces the old device plugin clunk for a cleaner, CEL-powered model that actually scales.

Big picture: Kubernetes gets smarter about where stuff runs. DRA pushes it toward topology-aware, parameter-tuned provisioning. Think tighter resource sharing, leaner clusters, and hardware that works for you—not the other way around.