Home
익명 06:28

Connect GPU to on-premise Kubernetes Cluster

Connect GPU to on-premise Kubernetes Cluster

I would like to deploy AI applications in my Kubernetes cluster.

Currently, my on-premise Kubernetes cluster has the following configuration:

  • 3 master nodes
  • 3 worker nodes

I was thinking of adding two local GPUs workers nodes equipped with GPU cards and scheduling the AI model pods on those nodes. However, I'm not sure if this is the cleanest, most scalable for deploying AI workloads in Kubernetes.

I know that NVIDIA provides a set of pulgin and drivers in kubernetes but is it the most suitable architecture and deployment strategy?


Top Answer/Comment:

Comment: As long as its not in a VM it should run just fine? That said the only time I tried it was with a microk8s cluster and yeah it wasn't a problem. Just base your images around a nvidia friendly docker container, and be sure to setup the node affinity so it ends up spinning on the right host. edit: Oh yeah: microk8s enable gpu , if it is a microk8s flavored cluster

상단 광고의 [X] 버튼을 누르면 내용이 보입니다