ResearchBenchmarksHardware Efficiency on GEMM M ≥ 128Hardware Efficiency on GEMM M ≥ 128Follow1.88SpeedupOSC1.58881.66441.741.8156Apr 14, 2026Evaluation resultsMethodMethodLinksSpeedupOSCGroup Size (G)=64Group Size (G)=642026.041.88OSCGroup Size (G)=32Group Size (G)=322026.041.78OSCGroup Size (G)=16Group Size (G)=162026.041.6