Computable GPU Index

Computable GPU Index is an open-source USD price index for GPU compute, built from published on-demand rental rates from a fixed provider panel. It helps traders, renters, financiers, and agents compare GPU pricing against a transparent reference rate.

AI交易工具

AI投資工具

AI分析助手

訪問網站

Overview

Computable GPU Index (CGI) is an open-source USD price index for GPU compute. It converts published on-demand rental rates from a fixed panel of providers into a single price per GPU-hour, giving the market a reference number that can be inspected rather than trusted as a black box.

The site frames the index as a transparent, reproducible alternative to closed pricing benchmarks. Computable says the method is mathematically robust, the data are kept with timestamps, and anyone can verify or recompute published prints from the public record.

Core capabilities

Frequent rate collection

Computable collects published on-demand rental rates every 15 minutes, so the index reflects a regularly refreshed snapshot rather than a static benchmark.

Fixed-provider screening

Only on-demand rates from a fixed panel of providers are included, which keeps the calculation anchored to a defined and repeatable data set.

Robust aggregation method

The index uses an interquantile mean over votes, with outer-third votes having no direct effect on the published value.

Source-backed records

Every data point is sourced, timestamped, and kept, which supports traceability back to the public record.

Public reproducibility

Users can clone the collector and the calculation from the repo and recompute a published print from the public record.

Multiple access paths

The live index is available through charts, history, a GitHub repo, and an MCP endpoint for agent use.

Practical use cases

Benchmark market pricing
Use the live index as a reference rate when comparing GPU rental quotes across providers or time periods.
Monitor price movement
Use the published history and live charts to track how prices change for specific GPU SKUs over time.
Reproduce the index
Clone the collector and calculation to verify a published print, or to audit how the value was derived from public records.
Agent consumption
Connect the MCP endpoint to an agent that needs a read-only, no-key source of current GPU pricing data.
Shared reference for compute decisions
Use the reference rate when trading, renting, or financing compute and you need a shared number instead of a provider-specific quote.

Pros and Cons

Pros

Open-source and public by design, with data, method, and code described as published.
Regularly refreshed from published on-demand rates collected every 15 minutes.
Uses a defined panel of providers and an interquantile mean to reduce tail influence.
Every published value is presented as reproducible from the public record.
Includes a live index, history, repo access, and an MCP endpoint for consumption by agents.

Cons

The source content is limited to the homepage, so broader workflow, setup, and deployment details are not documented here.
The site names the calculation method and access points, but does not publish a full methodology write-up in the provided material.
The pricing page URL returns a 404 in the collected sources, so commercial terms are not visible from the evidence provided.

FAQ

How is the GPU index calculated?

Computable says the index is built from published on-demand rental rates collected every 15 minutes from a fixed panel of providers. The published value is based on a mathematically defined method that the company says anyone can verify and reproduce.

Which GPU models are covered?

The live index currently lists H100, H200, B200, and B300 SKUs on the homepage. The site also says the index includes live charts and history for every SKU.

Can I reproduce the published number myself?

The homepage says the index is open by design: data, method, code, and every parameter are published. It also says you can clone the collector and calculation from the repo and recompute a published print.

Is there an API or agent-friendly interface?

The live index can be used directly in an MCP endpoint. The site describes it as anonymous, read-only, and requiring no key.

Who is this product for?

The homepage supports people who trade, rent, or finance compute and want a reference rate for GPU pricing.

Quick Facts

Category: GPU pricing index
Source domain: getcomputable.com
Primary output: USD price per GPU-hour
Method: Interquantile mean over votes
Update cadence: Every 15 minutes
Access: Live index, GitHub repo, MCP endpoint