-
@ExUtumno not sure if i grokked the BP/CP difference but I have a GPU version now thats 6x faster than my naive recursive python version on my GT640M laptop. It's super simple, basically just updating all tiles simultaneously until none of them change anymore.Permalink On twitter.com
❤️ 1 Favorite Mood +1 🙂
-
@ExUtumno to measure that I reduce the state of all of them to the sum of changes in the same pass, so the grid buffer doesn't need to leave the GPU even (except for visualization).Permalink On twitter.com
-
@ExUtumno Now if NVIDIA supported OpenCL2.0's device-side enqueue_kernel I could solve it completely on the GPU and there would be some smarter things I could try but eh ¯\_(ツ)_/¯On twitter.com