rackmount vs tower for homelab inference: worth it?

[labnest]

Thanks, that helps. I don't think your conclusion is far off.

Tried to reproduce it just now. My result was mixed: first run looked bad, second run looked normal. I expected the model choice to matter more than it did.