Computable GPU Index (CGI)

One trustworthy reference for something this opaque feels overdue to me, and being able to check how it's put together is what would make me actually believe it.

Congratulations on the launch team! Loved the launch video!!! :)

GPU economic almost as important as model economic . An independent index for tracking the cost of compute feels very timely.

Really like the transparency behind this. Having a GPU price index where the methadology is reproducible makes it much more useful than just another pricing dashboard.

Congrats on the launch! Do you include hyperscaler list prices in the provider panel?

Congrats on the launch from the Signify team! 🚀Tracking GPU pricing and performance across different providers is usually a mess, so having a clean benchmark index like CGI is a huge win for dev teams.Upvoted, and hoping you guys hit the top of the leaderboard today! 🙌

Are you planning to expand beyond H100/H200/B200/B300 at some point, maybe older or more niche chips too?

Awesome to see confirmability built in from day one instead of bolted on later. what actually triggered the idea, did you run into this exact pricing problem directly while working on Computable?

Finally, a GPU price index that doesn't make me feel like I'm deciphering ancient runes. Already sent this to my team, we've been manually tracking cloud costs in a spreadsheet like cavemen. Any chance you'll add Lambda Labs or Vast.ai to the provider list?

Hi Product Hunt, I'm Ray, cofounder of Computable.What is the price of a GPU? Nobody agrees. Every provider quotes a different number, and existing indexes are closed black boxes: a figure you are asked to trust without seeing the data, the method, or the code behind it. Despite being rented, resold, and financed at commodity scale, compute lacks the reference rate every mature commodity market has.CGI is built to be that number, and to earn trust in the open rather than by authority. It is open data, open method, open code.How it works: every 15 minutes we collect published on-demand rental rates from 28 providers. Each panel provider casts weighted votes at its price, and the index is an interquantile mean over those votes, not an average: only the central third of the vote mass is averaged. No small group of providers can drag the number from the tails.The weights are systematically scored. Liveness weights measure whether a provider contributes new information to the price: a leave-one-out ridge regression scores whether each provider's recent moves anticipated the rest of the panel, and the weights recompute every observation. Every weight is floored and capped, and no weight requires a per-provider human decision.The design rests on five properties: verifiable, reproducible, fault-tolerant, outlier-resistant, transparent. The data is verifiable, the method is transparent, the prints are reproducible via code, and the result is a robust price surviving data sources that disagree, err, or break.Live today: H100, H200, B200, B300. The collector and the calculation are open source. Clone the repo and recompute any print since inception; you'll get the same number we published.If you trade, rent, or finance compute, I'd love to hear what you want from a reference rate.