gpgpu-sim

Select a stat to view. One chart at a time.

Cycles normalised to real RTX 3060 (1.0 H100 speedup over RTX 3060 (both simulated), wall-clock time (cycle ratio x 1980/975 MHz) H100 speedup over RTX 3060 (both simulated), cycles only (clock removed) Share of simulated cycles spent in fixed kernel-launch latency L2 miss rate DRAM reads (requests) IPC (instructions per cycle, whole GPU) Largest kernel launch, CTAs (real 3060) Simulator wall time (s) H100 (simulated) speedup over my REAL laptop Real laptop: effective clock (GHz)