So far some basic research done with Claude:
https://github.com/jpietek/PenguinBurner/blob/main/docs/designs/2026-05-27-amd-gpu-backend-design.md
I can do a common interface for the auto-uv procedure and support AMD too, but the major pain point is that it seems hard to have per volatage point offset in RDNA 2+ cards as on Nvidia.
Worst case we will need to live with a fixed global offset in AMD universe.
So far some basic research done with Claude:
https://github.com/jpietek/PenguinBurner/blob/main/docs/designs/2026-05-27-amd-gpu-backend-design.md
I can do a common interface for the auto-uv procedure and support AMD too, but the major pain point is that it seems hard to have per volatage point offset in RDNA 2+ cards as on Nvidia.
Worst case we will need to live with a fixed global offset in AMD universe.