https://github.com/kubernetes/enhancements/issues/5075

Feature Overview (aka. Goal Summary)  

As an OpenShift admin, I want to split a single physical device (like a GPU or network card) into smaller pieces and give those pieces to different Pods so that I can use hardware more efficiently and avoid wasting resources.

What’s changing

• Today:

• • If a Pod uses a DRA ResourceClaim, it takes the whole device — even if it doesn’t need all of it.

• • Only way to share is if multiple Pods use the same claim (not flexible).

• New:

• • DRA drivers can mark a device as shareable.

• • Pods can each get their own claim for just part of the device (e.g., 4GB of GPU memory or 5Gbps of bandwidth).

Why it’s useful

• No more wasting a full GPU for a small workload.

• Different teams/namespaces can share the same hardware safely.

• Works for many resource types — GPUs, FPGAs, NICs, etc.