Get in touch

The best way to reach us is email:

hello@bitpute.com

We read everything. Replies usually go out within two or three working days, sometimes faster, occasionally slower if a message needs actual testing before we can answer properly.

What to write about

• A wrong number. This is the message we care about most. If a calculator's output disagrees with what you measured on real hardware, send the model, quantization, GPU, context length and, if you can, the nvidia-smi reading. We will re-check the formula and either fix it or explain the gap.
• A missing model or GPU. Tell us which one and we will look at adding it. Models with a public config file get added fastest, because we refuse to guess architecture numbers.
• A tool request. The roadmap on the homepage shows what's planned. If something you need isn't there, say so. Requests do change the order we build in.
• Privacy or legal matters. Requests under data protection law go to the same address, see the privacy policy for details on what we hold (very little).

What we can't help with

We don't do purchasing advice for specific budgets, and we can't debug your CUDA install. The calculators tell you what should fit and roughly what it costs to run; the rest is between you and your terminal.