Open Dataset: AI Compute Card Specification Library

MirrorFrog organizes the full chip specifications behind the site into a machine-readable open dataset, freely available for research, teaching, secondary development, and cross-comparison. The data is continuously updated with the site, covering mainstream and emerging AI accelerators from NVIDIA, AMD, Intel, Google TPU, AWS, Huawei Ascend, Cambricon, Moore Threads, and more.

Download

File	Description	Direct Link
chips.json	Full specs of 188 chips (architecture, process, compute, memory, bandwidth, TDP, etc.)	/chips.json
pricing.json	Reference pricing and currency per chip (some estimated / undisclosed, marked null)	/pricing.json

Note: Both files are auto-generated from the docs/cards/** Markdown sources by a prebuild script before npm run build; the source files are the source of truth for specs.

Field Documentation (chips.json)

Field	Type	Description
id	string	Unique chip identifier (used for comparison pages and links)
title	string	Display name
vendor	string	Vendor (nvidia / amd / huawei / cambricon / google …)
slug	string	Corresponding card page path, e.g. /docs/cards/nvidia/h100
description	string	One-line summary
keywords	string[]	SEO keywords
tdpW	number	Parsed TDP value (watts)
fp16Tflops	number | null	FP16 compute (parsed numeric; null if missing)
specs.release	string	Release date
specs.architecture	string	Architecture name
specs.process	string	Process node
specs.memory.capacity	string	Memory capacity
specs.memory.bandwidth	string	Memory bandwidth
specs.compute.fp32 / fp64 / fp8	string	Compute per precision (raw text)
specs.tdp	string	TDP raw text

License

The dataset is released under CC BY 4.0 (Attribution 4.0 International). You are free to copy, distribute, modify, and use it commercially, provided you give appropriate credit: MirrorFrog (https://mirrorfrog.com).

How to Cite

Text format:

MirrorFrog. AI Compute Card Dataset. https://mirrorfrog.com/docs/open-dataset

BibTeX:

@misc{mirrorfrog2026,
  title        = {AI Compute Card Dataset},
  author       = {MirrorFrog},
  year         = {2026},
  howpublished = {\url{https://mirrorfrog.com/docs/open-dataset}},
  note         = {CC BY 4.0}
}

Suggested Uses

• Cross-study research: Use chips.json for quantitative analysis of compute / energy efficiency / memory across generations.
• Tool integration: Plug the dataset into your own selector, table, or visualization.
• Redistribution: Embed this dataset into your product or report, with attribution.

If you spot data discrepancies, please submit a correction via the feedback entry on any site page.