Modern GPUs and AI accelerators are increasingly connected through scale-up
fabrics such as AMD xGMI today and emerging UALink systems [4]. Linux lacks
common, vendor-neutral infrastructure for reporting which accelerators are
directly connected, through which ports, and in what state.
Sorry for jumping in late to this discussion. We have hardware that uses
UALoE (https://www.amd.com/en/products/rackscale-solutions/helios.html).
The suggestion from Jason is sensible: a new subsystem for fabrics and
reuse ethernet stuff from networking subsystem where possible. We would
like to continue discussing this - Alejandro and I will be attending
Plumbers virtually.
Thanks,
Alex