Large symmetric eigenvalue problems are commonly observed in many disciplines such as Chemistry and Physics, and several libraries including cuSOLVERMp, MAGMA and ELPA support computing large eigenvalue decomposition on multi-GPU or multi-CPU-GPU hybrid architectures. However, these libraries do not provide satisfied performance that all of the libraries only utilize around 1.5\% of the peak multi-GPU performance. In this paper, we propose a pipelined two-stage eigenvalue decomposition algorithm instead of conventional subsequent algorithm with substantial optimizations. On an 8×A100 platform, our implementation surpasses state-of-the-art cuSOLVERMp and MAGMA baselines, delivering mean speedups of 5.74× and 6.59×, with better strong and weak scalability.
@article{arxiv.2511.16174,
title = {Pipelined Dense Symmetric Eigenvalue Decomposition on Multi-GPU Architectures},
author = {Hansheng Wang and Ruiyi Zhan and Dajun Huang and Xingchen Liu and Qiao Li and Hancong Duan and Dingwen Tao and Guangming Tan and Shaoshuai Zhang},
journal= {arXiv preprint arXiv:2511.16174},
year = {2025}
}
Comments
11 pages,16 figures. Our manuscript was submitted to the PPoPP'26 conference but was not accepted. The reviewers acknowledged it as a complete and solid piece of work; however, they noted that it lacks sufficient ablation studies