Source
- URL: https://x.com/guilhermeotina/status/2060495962192195599
- Author: Guilherme O’Tina (@guilhermeotina)
- Posted: 2026-05-29 22:58:14
Branch
1/ @guilhermeotina
the part that makes this tricky is the memory model. gpu wants massive flat uniform access, os wants virtualized per-process pages. youd need to either flatten the kernel into the gpu driver or give the gpu an mmu it doesnt want. interesting collision of two compute paradigms though