Dissecting the NVIDIA Hopper Architecture through Microbenchmarking and Multiple Level Analysis

This repository contains the code for benchmarking NVIDIA GPU performance. The relevant papers are as follows:

• Weile Luo, Ruibo Fan, Zeyu Li, Dayou Du, Qiang Wang, and Xiaowen Chu. "Benchmarking and Dissecting the Nvidia Hopper GPU Architecture." In 2024 IEEE International Parallel and Distributed Processing Symposium (IPDPS), pp. 656-667. IEEE, 2024.
• Weile Luo, Ruibo Fan, Zeyu Li, Dayou Du, Hongyuan Liu, Qiang Wang, and Xiaowen Chu. "Dissecting the NVIDIA Hopper Architecture through Microbenchmarking and Multiple Level Analysis." arXiv preprint arXiv:2501.12084 (2025).

If you find this work useful, please cite this project and our papers.

@inproceedings{luo2024benchmarking,
  title={Benchmarking and dissecting the nvidia hopper gpu architecture},
  author={Luo, Weile and Fan, Ruibo and Li, Zeyu and Du, Dayou and Wang, Qiang and Chu, Xiaowen},
  booktitle={2024 IEEE International Parallel and Distributed Processing Symposium (IPDPS)},
  pages={656--667},
  year={2024},
  organization={IEEE}
}

@article{luo2025dissecting,
  title={Dissecting the NVIDIA Hopper Architecture through Microbenchmarking and Multiple Level Analysis},
  author={Luo, Weile and Fan, Ruibo and Li, Zeyu and Du, Dayou and Liu, Hongyuan and Wang, Qiang and Chu, Xiaowen},
  journal={arXiv preprint arXiv:2501.12084},
  year={2025}
}

Recommended environment

• CUDA 12.6 or above
• Ubuntu 20.04

Build & Usage

In the folder, use make or ./compile.sh to build, and use ./run.sh or ./run_all.sh to run.

Acknowledgment

• https://github.com/shen203/GPU_Microbenchmark provides a reference for our regular unit tests.
• https://github.com/RRZE-HPC/gpu-benches provides a reference for our memory and TMA random access tests.
• We used the tools in https://github.com/blackjack2015/NV-DVFS-Benchmark to test the power consumption.