Publication

Had a couple of old GTX 1080 Tis lying around and wanted to see if I could get reasonable LLM decode speeds on them without Tensor Cores. Managed to get ~115 t/s on Qwen3-4B Open-sourced the project here if anyone's curious: t.co/VRuq94QCwU

@zxtheidiot Love this! Getting ~115 t/s on Qwen3-4B out of a couple of old GTX 1080 Tis with no Tensor Cores is such a fun way to give older hardware new life. Thanks for open-sourcing it! #LocalLLM #OpenSource