EUVIMEDEUVIMED entdecken
Crossref

Mixed-Precision SVD on GPUs via Ogita–Aishima Iterative Refinement

Angelika Schwarz, Rasmus Munk Larsen, Samuel Rodriguez, Lev Kruglyak, Christopher Baker

ACM Transactions on Mathematical Software · 29.10.2026

DOI: 10.1145/3860899 ↗

Kurz gefasst

Fachartikel in der Zeitschrift ACM Transactions on Mathematical Software, online erschienen am 29.10.2026. Verfasst von Angelika Schwarz und 4 weiteren Autor:innen. Die Arbeit nennt 31 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The Ogita–Aishima iterative refinement algorithm for SVD is based on matrix–matrix multiplication, making it an attractive approach for GPUs. We extend the algorithm to complex arithmetic and add cluster handling. The latter is based on Wedin's perturbation theory and a per-pair version of Ogita–Aishima's convergence analysis. Earlier implementations require well-separated singular values; the proposed cluster handling removes this restriction. Our implementation refines an FP32-quality SVD computed with cuSolver's cusolverDnXgesvdp to FP64-quality, while matching the accuracy of the corresponding FP64 routine on all test problems—twelve synthetic singular-value distributions and 38 matrices from the SuiteSparse collection. The performance depends on the spectrum and the FP32-to-FP64 ratio. Under Ozaki-II FP64 emulation on an NVIDIA RTX PRO 6000 Blackwell workstation GPU, which has a \(64\) : \(1\) FP32-to-FP64 ratio, the refinement is up to \(11.1\times\) faster than cusolverDnXgesvdp on spectra whose adjacent gaps are resolved by the FP32 SVD, and at most \(1.9\times\) slower on spectra for which they are not. The advantage is at most \(1.5\times\) on data-center GPUs with a \(2\) : \(1\) FP32-to-FP64 ratio, such as the NVIDIA B200. These results show the algorithm's potential on hardware with strong mixed-precision capabilities.

Zitieren

APA

Schwarz, A., Larsen, R. M., Rodriguez, S., Kruglyak, L., & Baker, C. (2026). Mixed-Precision SVD on GPUs via Ogita–Aishima Iterative Refinement. ACM Transactions on Mathematical Software. https://doi.org/10.1145/3860899

Vancouver

Schwarz A, Larsen RM, Rodriguez S, Kruglyak L, Baker C. Mixed-Precision SVD on GPUs via Ogita–Aishima Iterative Refinement. ACM Transactions on Mathematical Software. 2026. doi:10.1145/3860899

BibTeX
@article{Schwarz2026,
  author = {Schwarz, Angelika and Larsen, Rasmus Munk and Rodriguez, Samuel and Kruglyak, Lev and Baker, Christopher},
  title = {Mixed-Precision SVD on GPUs via Ogita–Aishima Iterative Refinement},
  journal = {ACM Transactions on Mathematical Software},
  year = {2026},
  doi = {10.1145/3860899}
}
Kostenlos starten →