{"name":"openTPU","repo":"FeSens/openTPU","url":"https://midorreal.com/project/fesens-opentpu","repository":"https://github.com/FeSens/openTPU","description":"An open-source AI accelerator, developed by AI: RTL, ISA, simulator, compiler and profiler in one repo. Runs Qwen3, LFM2.5 and Qwen3.5 on a Kintex-7 PCIe card.","facts":{"license":{"value":"Apache-2.0","source":"https://github.com/FeSens/openTPU","checked":"22 Nov 2026"},"language":{"value":"Verilog","source":"https://github.com/FeSens/openTPU","checked":"22 Nov 2026"},"releases":{"value":"0","source":"https://github.com/FeSens/openTPU/releases","checked":"21 Nov 2026"},"contributors":{"value":"3","source":"https://github.com/FeSens/openTPU/graphs/contributors","checked":"21 Nov 2026"},"archived":{"value":"no","source":"https://github.com/FeSens/openTPU","checked":"22 Nov 2026"}},"stars":1007.0,"writeup":{"what_it_is":"openTPU is an open-source AI accelerator project. It contains the hardware design (SystemVerilog), an instruction set, a bit-exact simulator, a kernel language with a compiler, and host software, all in one repository. The design runs on a Xilinx Kintex-7 PCIe card and is reported to run Qwen3, LFM2.5 and Qwen3.5 models.","audience":"Learners and engineers interested in how AI accelerators work, from kernels to hardware","claims":[{"kind":"specific","claim":"Runs ten modern models with their real weights on an Inspur YPCB-00338 card","excerpt":"runs ten modern models with their real weights on an Inspur YPCB-00338 card","status":"not_checked"},{"kind":"specific","claim":"The card produces the same tokens as the simulator, bit for bit","excerpt":"the card produces the same tokens as the simulator, bit for bit","status":"not_checked"},{"kind":"specific","claim":"Builds for the BCU1525 with a VU9P and four DDR4-2400 DIMMs","excerpt":"a VU9P with four DDR4-2400 DIMMs (76.8 GB/s at peak) and","status":"not_checked"},{"kind":"specific","claim":"Runs mixture-of-experts models larger than the card's 4 GiB by streaming experts from host storage","excerpt":"Mixture-of-experts models bigger than the card's 4 GiB run with their experts streamed from host","status":"not_checked"},{"kind":"broad","claim":"Described as a learning project that is easy to read end to end","excerpt":"this is a good place to start","status":"too_broad"}],"alternatives":[],"written_by":"AI, from the project's README","date":"2026-11-19"},"why_now":null,"owner_supplied":null,"score":{"verdict":"mid","hype":59,"reality":54,"gap":5,"momentum":0.6,"confidence":60,"version":"score-2.0","rule":"verdict-2.0","calculated_at":"2026-11-22T08:00:00+00:00","verdict_source":"rule"},"early_read":null,"changes":[{"date":"2026-09-24T12:26:10+00:00","text":"Repository created","source":"https://github.com/FeSens/openTPU"}],"cite":"openTPU is a GitHub project at https://github.com/FeSens/openTPU written mainly in Verilog, described there as \"An open-source AI accelerator, developed by AI: RTL, ISA, simulator, compiler and profiler in one repo. Runs Qwen3, LFM2.5 and Qwen3.5 on a Kintex-7 PCIe card\".","attribution":"Data from Mid or Real, https://midorreal.com/project/fesens-opentpu, checked 22 November 2026. Attribution with a link is required.","still_real":null}