Legal Professional Locally Deploys DeepSeek on Older V100 GPUs

On November 6, 2026, a post appeared in Reddit's LocalLLaMA community from a legal professional who runs a server built from older graphics cards. According to the post, the author had earlier tried running DeepSeek-V4.1-Flash on four V100s, before doing custom kernel work and prefill optimizations, and measured 36 to 38 tok/s at that point. He also said he would try it across eight cards. He then did so, deploying the model on eight V100s, which were manufactured in 2018. He reported that the results were, honestly and genuinely, good and useful to him in practice. In his testing, the setup delivered very high throughput and strong long-context performance on both coding tasks and legal writing. The author also made a point of saying that he did not use the model to paint a 3D island in blender, or to re-create a gameboy game in a web browser. No further updates on this deployment have appeared since then, so the current state of the project is this 8-card V100 setup running DeepSeek-V4.1-Flash with the custom kernel and prefill optimizations in place.