Windows GPU device plugin
author of the fork, 2022–2026
Where the upstream release groups a Windows machine’s GPUs into a single undifferentiated pool, this fork allows Kubernetes to identify and allocate each processor individually.
Kubernetes learns which GPUs a Windows machine has through a device plugin, and the upstream plugin reports them as a single pool. This fork exposes each GPU individually, reads the scheduler’s allocation requests accurately, and advertises a default number of devices.