Local test cluster - hardware requirements

I am trying to get an estimate on the total hardware count when putting together a GPU k8s cluster.

If I am budgeting for 4 GPU machines to serve as the nodes, is it reasonable/possible to use one of the node for what I understand are the control-plane or do I need to host that on one (or more?) nodes that are not part of the 4 GPU machines that are my nodes.


For a small cluster you could potentially host it on the same nodes - the one thing is you don’t want clash for resources with the control plane components. If you size the pods on it right it shouldn’t be too much of an issue. If you have $$ for it, a separate node (could be tiny for that size cluster) might be worthwhile.