NVIDIA Ships k8s-device-plugin v0.19.1 Fixing WSL and Tegra GPU Bugs

NVIDIA's k8s-device-plugin is the bridge between Kubernetes and your GPU hardware. Without it, Kubernetes has no idea your nodes have GPUs , it can't schedule GPU workloads, can't track device health, and can't enforce resource limits. The plugin has been the standard solution since 2017 and now sits at nearly 3,800 stars with over 800 forks. The latest release, v0.19.1, is a patch that ships two targeted bug fixes alongside a round of dependency updates.

What the plugin actually does

The plugin runs as a DaemonSet , a pod that Kubernetes automatically schedules on every node , and does three things: it exposes the number of GPUs on each node to the cluster scheduler, keeps track of GPU health, and enables GPU-enabled containers to run. Once deployed, pods can request GPUs using the standard nvidia.com/gpu resource type, just like they would request CPU or memory.

Here's the minimal pod spec to claim a single GPU:

resources:
  limits:
    nvidia.com/gpu: 1

The scheduler then places that pod on a node with a free GPU slot. No custom code, no driver wrangling inside your container , the plugin handles all the device injection at the runtime level.

What changed in v0.19.1

The v0.19.1 patch ships two functional fixes backported from the release-0.19 branch: a WSL fix that makes the plugin report a single "all" device to kubelet, and a correction to CDI spec generation to respect the driver root for Tegra CSV files. The rest of the changes are dependency bumps , Go runtime, gRPC, k8s.io libraries, and GitHub Actions versions.