Haven’t found answers to this yet searchign the forums - I’m trying to chain a number of Framework desktops together for distributed AI processing, but this is limited by only having two USB4 slots in the back. Are there any recommended/known-working USB-C/USB4 PCIe expansion cards people have used for the AI desktop?
Is this related at all?
What is the limitation your are facing? As the topologies available are not the same you’d expect on a switched network.
1 Like
@Squiggler that looks similar, I’ll take a deeper look thanks!
@Pedro_Leite limitation is I have 4 machines and 2 USB C ports. I have daisy-chained them, but the extra hops for the furthest machines adds latency. I’d ideally like a hub-and-spoke, with one center and three machines out, but … I need the extra port for that. Idea is that I can use RDMA and have low latency so I can do things like split-tensor processing, instead of the currently-supported split-by-layer processing