DDR5 Costs Five Times More. My 2013 Workstation Runs a 125B Model Anyway.

38 to 48 tokens per second from a Xeon without AVX2 and two 2017 GPUs, and what three home PCs taught me about where local LLM speed really comes from. The X79 up close: two used Titan Xp cards from 2017, the Xeon’s coo…