GC-MoE on SM120: Overlapping Expert-Parallel Token Dispatch with Warp-Level FP8xFP4 Block-Scaled MMA on Consumer Blackwell
AI-generated summary A published paper presents an expert-parallel Mixture-of-Experts decoding operator rebuilt for consumer Blackwell SM120 hardware over PCIe. The implementation achieves performance improvements over FlashInfer CuTe DSL W4A8 MegaMoE and Cake MegaMoE across evaluated decode points.