OpenTPU: Open-source AI accelerator with RTL, compiler and tools

OpenTPU is an open-source AI accelerator project that provides a complete stack: register-transfer-level hardware design, instruction set architecture, simulator, compiler and profiler in a single repository. The project demonstrates real functionality by running Qwen3, Llama 2.5 and Qwen 3.5 models on a Kintex-7 FPGA PCIe card, making the full design accessible to developers and researchers.

The release bundles both the theoretical and practical components needed to understand and build AI accelerators, lowering barriers for hardware-focused AI work outside major chip manufacturers.