RunsOnMy ModelsGPUs & MacsAPI pricesVRAM calculatorNew

LLM VRAM calculator

Pick a model and your GPU or Mac. The maths runs in your browser with the same code that builds every page on this site.

Loading data…

How it works

Total memory = weights (real GGUF file size, or parameters × bits per weight) + KV cache (bytes per token from the model's config × context) + overhead (0.4 GiB runtime + logits buffer). A model "fits" when the total is under 90% of usable memory, and is "tight" up to 100%. Full details on the methodology page.