Contact

The quickest way to reach us is email: hello@vramcalculator.net. We read every message and usually reply within two to three working days.

Report a wrong number

Corrections are the most useful messages we get. To help us check quickly, please include:

  • The page or calculator you used, and the model and GPU.
  • The quantization (for example Q4_K_M) and the context length you set.
  • Your runtime and version (llama.cpp, Ollama, LM Studio, vLLM, ExLlama, ComfyUI) and your operating system.
  • The memory or speed your runtime reported, ideally a screenshot or log line such as the output of ollama ps or nvidia-smi.

Small differences of a few hundred megabytes are normal between runtimes. Differences of 10% or more usually mean a setting we should document or a formula we should improve.

Request a model or GPU

Send a link to the model on Hugging Face, or to the GPU's official specification page. We add models that have a public config.json, and GPUs and systems with published memory size and bandwidth.

Using our data

You are welcome to quote our figures and use the public JSON data with attribution and a link to the page you used. If you are building something larger on top of it, tell us; we are happy to help.

Partnerships and advertising

For partnership or advertising enquiries, email us with a short description. We do not sell placement in comparisons, rankings or guides, and sponsored content, if we ever publish any, will be clearly labelled.

Privacy

We only use the details you send to reply to you. See the privacy policy for how the site handles analytics and advertising.