Sources & acknowledgments

Built on the work of a generous community.

Thank you

AMD-Pro-Bench exists because people test their hardware, document their setups and share what they learn. Thank you to the benchmark authors, independent reviewers, community contributors, model creators and open-source developers who make this collection possible.

We are especially grateful to the contributors behind BetterBench, llama.cpp, vLLM, ComfyUI, SD.Next, stable-diffusion.cpp, ROCm, and the many other harnesses, engines and workflows linked in our reports. Original work retains its attribution and licensing; every benchmark links back to its source.

Measured results, traceable sources

Our benchmark collection is based on documented, source-reported measurements. We do not use performance predictions, theoretical GPU peaks or unsupported claims as measured benchmarks. Reposts and cross-listed results are checked for duplication.

Any source-derived estimates or calculated metrics retained for context are explicitly labeled and kept distinct from measured results. Documentation quality varies, so source strength, missing details and caveats remain visible. Source review does not mean we independently reproduced a result.

Source websites

Made possible by you

Thank you to everyone who shares a result, improves an open-source tool or helps make AMD AI performance easier to understand. If you or your project have been left off this list, or a credit needs correcting, please let us know in the community forum. We would be glad to recognize your contribution.