It'll be less than a networked cluster, but it still mattered with Threadripper units and I'd expect a racked board of this nature to expose more disparity between accessing memory in other chiplet areas.