Data sources & attributions
Each row on HWBench links to its exact source. Third-party data stays under the licence or terms of its source, as listed below. HWBench's own compilation, derived metrics and evidence grades are licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). See the Terms of use at /terms.html.
Where a source is a website rather than an openly licensed repository, we reproduce only individual factual figures with a citation and link. We never copy text, tables in bulk or images.
| Source | Used for | Licence / terms | How we use it |
|---|---|---|---|
| MLCommons MLPerf Inference results v3.1, v4.0, v5.0, v5.1, v6.0 and v6.1, https://github.com/mlcommons/inference_results_v6.1 (and the matching repositories for each version) | Audited datacenter and edge results | Apache License 2.0, as published in each results repository | Figures cited per result. MLPerf is a trademark of MLCommons; see the Disclaimer |
| Hailo Model Zoo v2.20.0, https://github.com/hailo-ai/hailo_model_zoo | Hailo-8, 8L and 10H model throughput | MIT License | Figures cited per model |
| Qualcomm AI Hub Models, https://github.com/qualcomm/ai-hub-models | Snapdragon and QCS latency | BSD 3-Clause License | Figures cited per model. No endorsement by Qualcomm is implied |
| AMD Vitis AI, https://github.com/Xilinx/Vitis-AI | FPGA and Versal model performance | Apache License 2.0, as published in the repository | Figures cited per model |
| llama.cpp, https://github.com/ggml-org/llama.cpp | LLM tokens per second | MIT License | Figures cited per result |
| OpenVINO, https://github.com/openvinotoolkit/openvino | Intel benchmarks | Apache License 2.0 | Figures cited per result |
| Seeed Studio wiki-documents, https://github.com/Seeed-Studio/wiki-documents | Jetson and reComputer figures | Licence as stated in the repository | Individual figures with citation only |
| Ultralytics documentation, https://docs.ultralytics.com/ | YOLO benchmark figures | Ultralytics website terms. The Ultralytics software is AGPL-3.0, but HWBench does not use or distribute that software | Individual figures with citation only. No code or text copied |
| NVIDIA Jetson benchmarks, https://developer.nvidia.com/embedded/jetson-benchmarks | Jetson figures | NVIDIA website terms of use | Individual figures with citation only |
| AMD and Xilinx documentation, product briefs and technical articles (amd.com, docs.amd.com, xilinx.com) | Versal, Kria and Ryzen AI figures | AMD website terms of use | Individual figures with citation only |
| strixhaloguide.com, vminstall.com | Strix Halo, Apple and Jetson Thor comparisons | Website terms of each site | Individual figures with citation only. Graded low-evidence (D/E) |
Licence notices
- Apache License 2.0 (MLCommons results, Vitis AI, OpenVINO): https://www.apache.org/licenses/LICENSE-2.0. Copyright belongs to the respective contributors as stated in each repository's LICENSE and NOTICE files.
- MIT License (Hailo Model Zoo, llama.cpp): https://opensource.org/license/mit. Copyright belongs to Hailo Technologies Ltd. and to the ggml authors respectively, as stated in each repository's LICENSE file. The MIT permission notice applies: the software and data are provided "as is", without warranty of any kind.
- BSD 3-Clause License (Qualcomm AI Hub Models): https://opensource.org/license/bsd-3-clause. Copyright belongs to Qualcomm Technologies, Inc. and/or its subsidiaries, as stated in the repository's LICENSE file. The BSD-3 disclaimer applies, and the names of the copyright holder and its contributors are not used to endorse HWBench.
- CC BY 4.0 (HWBench's own compilation): https://creativecommons.org/licenses/by/4.0/
To report a missing or incorrect attribution, email [email protected].
Fonts
| Font | Copyright | Licence |
|---|---|---|
| IBM Plex Sans, IBM Plex Mono, IBM Plex Sans Hebrew | © 2017 IBM Corp., Reserved Font Name “Plex” | SIL Open Font License 1.1 (unmodified font files, self-hosted) |
| Atkinson Hyperlegible (readable-font option) | © 2020 Braille Institute of America, Inc. | SIL Open Font License 1.1 (unmodified, self-hosted) |
Platform photos
Photos are stored on this site (not hotlinked). Platforms without a licensed photo show a neutral line drawing. Machine-readable list: /images.json.
| Image | Shows | Used for | Author | Licence | Source |
|---|---|---|---|---|---|
 | Raspberry Pi 5 | Raspberry Pi 5 (CPU only) | SimonWaldherr | CC BY 4.0 Resized by Wikimedia Commons thumbnailer (960 px wide); no other changes. | Wikimedia Commons file page |
 | Raspberry Pi AI HAT+ (Hailo-8) | Raspberry Pi 5 + AI HAT+ (Hailo-8) | SparkFun Electronics | CC BY 2.0 Resized by Wikimedia Commons thumbnailer (960 px wide); no other changes. | Wikimedia Commons file page |
 | NVIDIA Jetson AGX Orin developer kit | NVIDIA Jetson AGX Orin 64GB NVIDIA Jetson AGX Orin (MaxQ) | Auledas | CC BY 4.0 Resized by Wikimedia Commons thumbnailer (960 px wide); no other changes. | Wikimedia Commons file page |
 | NVIDIA Jetson Orin NX module on a carrier board | NVIDIA Jetson Orin NX 16GB NVIDIA Jetson Orin NX 16GB (MaxQ) | 4300streetcar | CC BY 4.0 Resized by Wikimedia Commons thumbnailer (960 px wide); no other changes. | Wikimedia Commons file page |
 | NVIDIA DGX Spark | NVIDIA DGX Spark | Daniel Lu (User:dllu) | CC BY-SA 4.0 Resized by Wikimedia Commons thumbnailer (960 px wide); no other changes. | Wikimedia Commons file page |
 | Apple Mac mini (M4, 2024) — the M4 Pro model shares this enclosure | Apple M4 (10-core GPU, 120 GB/s) Apple Mac mini M4 Pro 64GB | Seasider53 | CC BY 4.0 Resized by Wikimedia Commons thumbnailer (960 px wide); no other changes. | Wikimedia Commons file page |
Machine-readable: /find.json (see llms.txt); full tables are in “Show data”. Confidence grades: A audited MLPerf closed · B documented third-party · C community · D vendor-published · E estimate. Unknown values are shown as unknown, never estimated. Generated 2026-10-16 from public sources.