Image: github.com · GitHub repository card for LeelaChessZero/lc0
Worked Proof shown
LeelaChessZero's CUDA backend ported to AMD ROCm
Claude Code ported the chess engine's GPU backend so it runs on AMD cards.
—
Credits
- Repository account LeelaChessZero Source ↗
- HN submitter CalChris Source ↗
Earliest dated source:
The full story
This pull request to Leela Chess Zero, the open-source neural network chess engine, adds a complete ROCm backend so it runs on AMD GPUs, including modern attention-based networks on RDNA 3.5. It was ported from the CUDA backend with Claude Code. It uses rocBLAS and MIOpen, comes in FP32, FP16 and auto-detect variants, and detects the GPU architecture automatically.
How it was done
Tools
Themes
Loading themes…
Sources
From the maker's own write-up or code. We summarise what they report; we haven't tested it ourselves.
- Hacker News: Claude Code Ported LeelaChessZero CUDA Back End to ROCm: End of CUDA MoatRepository account: LeelaChessZero
The pull request with implementation details.
- Hacker News discussion (2 points)HN submitter: CalChrisPosted