ResearchBenchmarksHardware Efficiency on GEMM M = 16-64Follow1.72SpeedupOSC1.49121.55061.611.6694Apr 14, 2026Evaluation ResultsMethodMethodLinksSpeedupOSCGroup Size (G)=64Group Size (G)=642026.041.72OSCGroup Size (G)=32Group Size (G)=322026.041.64OSCGroup Size (G)=16Group Size (G)=162026.041.5