About RigsForLLM

Buying hardware to run language models at home mostly comes down to two numbers: how much memory the GPU can use, which decides what fits, and how fast that memory moves, which decides how quickly the answer arrives. Spec sheets print neither in a form you can compare across a Mac, a Ryzen AI Max box and a pair of graphics cards.

This site puts 24 machines you can order and 12 DIY builds against each other, checks each against 41 open models, and shows the token speeds people have published on the same chip or cards, with a link to every run. Where nobody has measured, the figure is an estimate and says so. Where a maker does not publish how much memory the GPU can use, the figure is an assumption and says so.

There is no advertising, no shop links and no sponsored placement. Prices are maker list prices, rounded and approximate.

The full method is on how we compare. Corrections, missing machines and benchmark runs we should include are welcome through the contact form.