Before I go all in on Jetson Orin Nano cluster for local llm, I’m wondering if anyone has tried this clustering using something other than the onboard NIC?
I’d like to think an upgraded interconnect would be of massive performance benefit, but am wondering to what extent the CPU can handle it while running models?
I’m thinking a USB->2.5Gb or USB->5Gb dongle?
or possibly even an M.2 → 10Gb ?