V100

Ten-Year-Old V100 Runs LLMs Again — AI Inference Cost Is Underestimated

A Reddit user ran 35B LLMs on Nvidia's 2017 V100 via a vLLM fork, matching new consumer chips. AI inference costs may be looser than most assume.