About this site
A calculator for one question: will this model run on my machine, and how well?
Downloading a local model is a forty-gigabyte commitment made on the strength of a forum post. The advice you find is either a rule of thumb that ignores context length, or a table that treats every model of the same size as identical. Neither survives contact with a real machine, and both cost people an evening and a lot of bandwidth.
So this site does the arithmetic properly. It reads each model's own architecture — layer count, attention head grouping, expert routing, cache design — and works out what that specific model needs on your specific hardware, at the context length you actually intend to use. 133 model architectures, 131 hardware profiles, and a formula for every number, written out in full so you can check it.
How it stays current
The catalogue is rebuilt from the Hugging Face index rather than maintained by hand. When a model is published and people start using it, it appears here on the next refresh with its real configuration attached. Nobody has to notice it and type it in, which is the only way a catalogue like this stays honest for more than a month.
How it is paid for
Advertising, and Amazon affiliate links on the hardware upgrade suggestions. If you buy a card through one of those links we earn a commission and you pay the same price. We do not take payment for placement, no vendor has any say in the rankings, and the upgrade tables are ordered by memory capacity — a number we did not choose — rather than by what pays best.
Where a recommendation would be bad advice, we say so even when the alternative is a cheaper purchase. The tables refuse to recommend a quantisation low enough to break a model, for instance, which routinely means telling someone their existing card is fine rather than pointing them at a bigger one.
What we do not do
- No accounts, no sign-up, no newsletter.
- No tracking of what you check. The calculator runs entirely in your browser; your hardware choice is stored locally so you do not have to pick it twice, and is never sent anywhere.
- No benchmark claims. We publish estimates and label them as estimates.
- No model quality rankings. Plenty of places will argue about which model is smarter. We answer the narrower question of whether it fits.
Getting in touch
Corrections are welcome, particularly measured numbers that disagree with ours — that is how the estimates get better. Missing hardware and missing models are worth reporting too. The contact page has the details.