GPUCostLab

About GPUCostLab and its author

Who writes GPUCostLab, the credentials behind it, how the site is funded, and how to report a wrong price or specification.

By Updated

The author

GPUCostLab is written and maintained by one person, a Principal ML engineer, who publishes under the site's name.

Why this site

I contribute to vLLM, the inference engine whose memory model the VRAM calculator follows. I created Videoflow, and I was the lead ML platform engineer for a 5,000-stream computer-vision platform. GPUCostLab applies that background to a question people ask before they rent a GPU: will the model fit, and will it cost less than an API?

The site exists because most answers to that question skip the arithmetic. A model "needs 16 GB" until someone asks at what context length and with how many concurrent requests. A GPU is "cheapest" until you divide by the throughput you actually get. GPUCostLab shows the formula, links every price to the page it came from, and says when a number is an assumption.

How the site is funded

The calculators and the price table are free. The site may earn referral or affiliate commissions from some GPU providers and, on the home-hardware guides, from retailers such as Amazon. Every such link will be labeled "(paid link)" where it appears. Today none are active, so every vendor link is a plain link. The affiliate disclosure lists every relationship and the rules that keep commissions out of the numbers. There is no advertising and no sponsored content.

Independence

GPUCostLab is not affiliated with NVIDIA, AMD, any cloud or API provider, or any model publisher. Providers cannot pay for placement or ranking, or to be left out. The site publishes no reviews or testimonials.

Contact

Email [email protected] to report a wrong price, specification or model shape, or a provider page that has changed. Corrections are dated on the page they affect.

Also available as Markdown.