| license: mit | |
| tags: | |
| - triton | |
| - windows | |
| - nvidia | |
| - gpu | |
| - cuda | |
| - compiler | |
| # Triton for Windows (NVIDIA) | |
| **Pre-built Windows wheels for Triton - NVIDIA GPUs only** | |
| ## Author | |
| **Dr. Ankush Rai** | |
| - [GitHub](https://github.com/RaiAnk) | |
| - [LinkedIn](https://www.linkedin.com/in/dr-ankush-rai-80722021a/) | |
| - [YouTube](https://www.youtube.com/@AnkushRai_polymath) | |
| ## Source Code | |
| **GitHub:** [https://github.com/RaiAnk/triton-windows-nvidia](https://github.com/RaiAnk/triton-windows-nvidia) | |
| ## Installation | |
| ```bash | |
| pip install https://huggingface.co/raipolymath/triton-windows/resolve/main/triton-3.6.0+git84bd6d54-cp311-cp311-win_amd64.whl | |
| ``` | |
| ## Requirements | |
| - Windows 10/11 (64-bit) | |
| - Python 3.11 | |
| - NVIDIA GPU (Compute Capability 8.0+) | |
| - CUDA Toolkit 11.8+ | |
| ## About Triton | |
| Triton was originally created by [Philippe Tillet](https://github.com/ptillet) as a research project at **Harvard University**. The project gained significant momentum when he joined OpenAI, who released it as an open-source Python-based DSL in 2021. | |
| This Windows port focuses on **NVIDIA GPUs only** (AMD backend disabled due to LLVM compatibility issues on Windows). | |
| ## Acknowledgments | |
| - **Philippe Tillet** - Creator of Triton (Harvard University) | |
| - **OpenAI** - For open-sourcing and continued development | |
| - **LLVM Project** - Compiler infrastructure | |
| - **NVIDIA** - CUDA toolkit | |