This is truly awesome. I love seeing the work done in this space. An on premise baby cluster all in 2U is amazing. I wonder what the commercial opportunities are for something like this…pre-installed with K8s, fully networked via the backplane, storage-less or storage-full, etc. A single 10Gb link out for public access and / or cluster access to external storage. Speed up the cluster? Just decommission a node, slide the latest compute module in and provision it from the management control plane. I think I’d buy something like that.
I really like the idea of a dedicated GPU when docked. Add a backup M.2 and it’s a winner. It has to “just work” though, that’s the major milestone.