ROCm/k8s-device-plugin

What it solves

This project provides a way for Kubernetes clusters to recognize and use AMD GPUs. Without a device plugin, Kubernetes cannot natively schedule compute workloads that require specific hardware acceleration from AMD GPUs.

How it works

It implements the Kubernetes device plugin interface, allowing AMD GPUs to be registered as schedulable resources within a container cluster. It is typically deployed as a DaemonSet across all nodes equipped with AMD GPUs, ensuring that the hardware is exposed to the cluster's scheduler. It can also be integrated with the AMD GPU Operator for broader resource management.

Who it’s for

System administrators and DevOps engineers managing Kubernetes clusters that use AMD hardware for compute-intensive tasks, such as machine learning or high-performance computing.

Highlights

• Kubernetes Integration: Enables AMD GPUs to be treated as standard schedulable resources.
• Health Monitoring: Supports experimental granular health detection per GPU via a gRPC socket service.
• Flexible Deployment: Can be deployed via standard YAML manifests or a Helm Chart.
• Node Labelling: Includes a node labeller to add additional GPU properties to cluster nodes.