About VRAMNeeded

VRAMNeeded answers one question per page: how much GPU memory does this AI model need to run on your own hardware? It's for anyone deciding whether to download a model, buy a GPU or rent one.

How we get our numbers

• Model facts (parameter count, layers, context length) come from the official model card and config.json.
• Memory per quantization is the real file size of each download on Hugging Face.
• KV cache is calculated from the config with the formula shown on each page.
• GPU and Mac memory come from NVIDIA and Apple spec pages.
• Rental prices come from the provider's public pricing page on the "checked on" date. Prices change often, so always confirm before you rent.

If a figure can't be traced to a source, it isn't on the site.

Corrections

Spotted a wrong number? We fix errors as soon as we confirm them, and we update the page's "checked on" date when we do.