#1750 Multi-node support for dual Strix Halo setups?

Beschreibung

Hi! I noticed the multi-GPU support and was wondering whether Strata could also run a model across two separate Strix Halo machines, using the memory on both to run higher-precision quantizations. I’m also interested in whether this could work with other systems that have large amounts of unified memory.

Is this already possible, or is multi-node support something you’re planning to add?

If it isn’t supported yet, how involved would it be to implement in a personal fork with help from AI coding tools? Could the existing multi-GPU implementation be a useful starting point, or would this require major architectural changes?

Any pointers would be appreciated. Thanks!