Conversations about the newest models, the hardware that runs them, and doing it yourself, from the people who run a rack of 90 used GPUs. The first episode is on its way. Sign up and get it the day it drops, plus our lab notes.
One email when there's a new episode or lab note. No spam, unsubscribe anytime.
Short write-ups from the rack: what came out, whether it runs on used hardware, and what broke. The longer story is on the show.
Qwen3.6-35B-A3B on Tesla P100s, RTX A4500s and ex-mining CMP cards. The 2016 P100s never throttled, and the A4500s won every speed test while sitting on their power cap.
A 74B mixture-of-experts model with 9B active, trained to run at 2-bit, with vision built in. Open weights, commercial use allowed.
From your first model on the computer you already own to a server that runs your whole team. Every step has guides in Learn.
Install Ollama, then run one command. Works on most laptops.
ollama run qwen3:8bMemory, quantization, and why one card is fast and another isn't.
Match a budget to the models you want. Use the build finder.
Parts, drivers, cooling, and the cables nobody tells you about.
Ollama, then llama.cpp, then Open WebUI for a ChatGPT-style app.
Chat with your documents, run agents, add voice. Or try it now in the Studio.