Source
- URL: https://x.com/steffenbauch/status/2005257271370977283
- Author: Steffen Bauch (@steffenbauch)
- Posted: 2025-12-28 12:39:24
Branch
1/ @steffenbauch
@SebAaltonen
I think this part sounds self-contradictory:
CUDA has a simple design for GPU memory allocation: The GPU malloc API takes the size as input and returns a mapped CPU pointer. The GPU free API frees the memory. CUDA doesn’t support CPU mapped GPU memory.
Is there a typo?