GPU clusters can deliver substantial compute resources, but coordinating communication between accelerators is a central design challenge. Michael LeBeane’s work examines GPU-initiated networking, software and hardware approaches, and performance modeling—topics relevant to engineers evaluating scalable computing systems.