DeepSeek-V4.1-Flash
Eight Old GPUs Run a High-Speed LLM, Rewriting the Bar for Local AI
A legal developer ran DeepSeek-V4.1-Flash on eight 2018 V100 GPUs at 215 tokens/s, highlighting practical local AI on retired hardware.
Back to home
V100
4 articles tagged with this topic
Yandex
80B Model Post-Trained on 3 GPUs—Dev Calls V1 'Basically Useless'
Developer post-trains Yandex's 80B model on 3 V100 GPUs, calls v1 'basically useless.' Training barriers are falling fast, but the runnable-to-usable
V100
$5,400 eBay V100 Rig Hits 200 Tokens/Sec on 27B — Old GPUs Power Local LLM Lab
$5,400 buys 8 used V100s hitting 200 tokens/sec on 27B models. Old GPUs + open-source tools are crushing local LLM hardware costs.
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.