Computable GPU Index (CGI) provides a transparent, open-source reference rate for GPU compute by aggregating on-demand rental prices into a single, verifiable USD value.
What the product does and how it is positioned
Computable GPU Index (CGI) serves as a standardized reference rate for the GPU compute market, providing a USD price per GPU-hour for various hardware models.
The system aggregates published on-demand rental rates from a fixed panel of providers to generate a single, mathematically robust index value every 15 minutes.
Designed for transparency and reliability, the index allows anyone to verify and reproduce published figures using open-source code and historical data records.
Source-supported ways to use the product
Users can track historical price trends for specific GPU models like the H100 or B200 to inform procurement and budgeting decisions.
Developers can connect AI agents to the live index via the Model Context Protocol to provide real-time compute pricing data for automated workflows.
The documented workflow, where available
The system automatically collects published on-demand rental rates from a fixed panel of providers every 15 minutes.
The pipeline filters the collected data to ensure only on-demand rates from the designated panel are included in the calculation.
Each provider in the panel casts weighted votes based on their current published rental prices.
The system calculates an interquantile mean, averaging the central band of votes while excluding the outer thirds to help reduce outliers from skewing the result.
The CGI utilizes an interquantile mean to ensure the resulting price index remains resistant to outliers and potential manipulation. By focusing on the central band of provider votes, the methodology is intended to reduce a small group of providers from significantly dragging the index toward extreme highs or lows.
This approach is designed to be fault-tolerant; even if a specific data source becomes unavailable or reports incorrect data, the index remains stable due to its reliance on a fixed panel and the exclusion of tail-end values.
Checks to run with your own material and workflow
What was checked and when
Answers based on the source-checked product record
The index collects published on-demand rental rates and updates its values every 15 minutes to reflect current market conditions.
The index currently tracks several high-performance SKUs, including the H100, H200, B200, and B300.
Yes, the methodology and code are open-source, allowing anyone to recompute any published value using the public record and the provided repository.
The index exclusively uses published on-demand rental rates from a fixed panel of providers, rather than spot or reserved pricing.
No, the Model Context Protocol (MCP) integration is anonymous, read-only, and does not require a key for access.
