← Journal · · Note
Native GPU resource support in K8s
Hey everyone! Look how interesting GPU support in K8s is becoming. I told you earlier about a new feature, DRA (Dynamic Resource Allocation) in Kubernetes. Well, yesterday at KubeCon they gave a detailed talk about this feature in the context of GPUs :)
We're in for an interesting new way to define GPU resources with expanded capabilities :)
Starting from version 30, native GPU sharing via time-slicing and MPS is already available.
From version 33, dynamic MIG will be available (right now it's only possible with workaround-style hacks).
Of course I'll share research on how this feature can help us with GPU management in K8s, how it differs from current approaches, and what possibilities it opens up. So stay tuned for updates 🤩