服务

律师的NVIDIA V100 AI服务器本地运行DeepSeek,升级至1TB内存并实现接近NVFP4的精度

来源标题:Lawyer’s NVIDIA V100 AI Server Runs DeepSeek Locally, Upgraded With 1 TB RAM And Near-NVFP4 Precision

A lawyer has assembled an unusually powerful private AI server using eight NVIDIA Tesla V100 GPUs from 2018, providing 256 GB of combined HBM2 memory to run DeepSeek V4.1-Flash locally. The setup originally achieved an impressive 215 tokens per second in code generation, but its creator has since upgraded the system with considerably more RAM, improved quantization, and additional hardware. The project belongs to Nick Mermiges, a trial lawyer and founder of PLI Labs, who detailed his custom-built system on Reddit, before publishing a follow-up update on November 12. His primary motivation is…

来源 wccftech.com

报道

  1. 律师的NVIDIA V100 AI服务器本地运行DeepSeek,升级至1TB内存并实现接近NVFP4的精度 Lawyer’s NVIDIA V100 AI Server Runs DeepSeek Locally, Upgraded With 1 TB RAM And Near-NVFP4 Precision wccftech.com