EXT64
Splendid
At set intervals (I can't remember what it usually is, perhaps every 2%?) the core checks the current result for errors. When it does this the GPU goes idle for a short period of time.
There have actually been at least 3 generations of OpenMM cores. The first was the AMD Core 16, now mostly retired. Then Core 17, the OpenCL core for both AMD and NVidia (used OpenMM 5.1). And now the ocores, which use the new streaming work model and OpenMM 6.
There have actually been at least 3 generations of OpenMM cores. The first was the AMD Core 16, now mostly retired. Then Core 17, the OpenCL core for both AMD and NVidia (used OpenMM 5.1). And now the ocores, which use the new streaming work model and OpenMM 6.