4 ms·

There is a PR[1] to extend the current alpha GPU implentation. Hopefully it will get merged soon. [1] - https://github.com/kubernetes/kubernetes/pull/28216 htt

by eudoxus 10y ago

There is a PR[1] to extend the current alpha GPU implentation. Hopefully it will get merged soon. [1] - https://github.com/kubernetes/kubernetes/pull/28216 https://github.com/kubernetes/kubernetes/pull/28216

puzzle 10y ago
That's still NVIDIA-only, plus you can't even tell a P100 from a K80 for scheduling purposes.
- boulos 10y ago
  I'd just use a separate NodePool and match your jobs using a selector as appropriate.
  - puzzle 10y ago
    So, is gci going to come with the right drivers?
- nanliu 10y ago
  This is the reason node-feature-discovery exists. Kubernetes needs an extensible way to detect and label hardware features: See: https://github.com/kubernetes-incubator/node-feature-discovery https://github.com/kubernetes-incubator/node-feature-discove...