Preprint
Sep 2026
VarioPath: Workload-Aware All-to-All Communication for PCIe GPU Clusters
AlltoAllv communication is a critical primitive in distributed large-model inference, particularly for mixture-of-experts (MoE) models. The growing adoption of PCIe GPU systems for cost-efficient inference makes AlltoAllv performance on these systems increasingly important. Without a dedicated scale-up interconnect (e....
Yao Fei, Jin Fang, Si-Ze Zheng et al.
· 0 citations