COMPLETE HARDWARE + SOFTWARE SETUP FOR RUNNING DEEPSEEK-R1 LOCALLY

Summary from the TLDR newsletter

This thread details a $6,000 setup for running the full DeepSeek-R1 model locally. The setup requires two CPUs and 768GB of RAM across 24 RAM channels. It doesn't include a GPU, though a GPU can be added for faster generation speed. The build described in the thread can generate six to eight tokens per second, depending on the specific CPU and RAM speed purchased. A video of the build running inference is available in the thread.