Advanced Computing Platform for Theoretical Physics

Skip to content
follow test15_5, 1. 32^4统一格子MG基线套件+fatbin sm_60/70+一一对应缓存,V100 8^4 1.098x vs L1 (2.25x vs BiStabCG) 32^4 L1 3.05s 2L OOM已定位;[opencode].