GPU Based Acceleration of Parallel Block Lancoz Solver
- Authors: Zamarashkin N.L.1, Zheltkov D.A.1
-
Affiliations:
- Institute of Numerical Mathematics of RAS
- Issue: Vol 39, No 4 (2018)
- Pages: 596-602
- Section: Article
- URL: https://journals.rcsi.science/1995-0802/article/view/202106
- DOI: https://doi.org/10.1134/S1995080218040169
- ID: 202106
Cite item
Abstract
The block Lanczos method for huge sparse linear systems over large prime finite fields is accelerated on GPU. Calculations on GPU are used for the operations with the dense matrices and blocks. The achieved acceleration of block operations significantly increases the parallel resource of the entire method, expanding the scalability area close to linear. The numerical experiments were carried out on supercomputers Lomonosov and Lomonosov-2.
About the authors
N. L. Zamarashkin
Institute of Numerical Mathematics of RAS
Author for correspondence.
Email: nikolai.zamarashkin@gmail.com
Russian Federation, Gubkina str. 8, Moscow, 119333
D. A. Zheltkov
Institute of Numerical Mathematics of RAS
Email: nikolai.zamarashkin@gmail.com
Russian Federation, Gubkina str. 8, Moscow, 119333