Benchmarks of Cuda-Based GMRES Solver for Toeplitz and Hankel Matrices and Applications to Topology Optimization of Photonic Components

Generalized Minimal Residual Method (GMRES) was benchmarked on many types of GPUs for solving linear systems based on dense and sparse matrices. However, there are still no GMRES implementation benchmarks on Tesla V100 compared to GTX 1080 Ti ones or even for Toeplitz-like matrices. The introduced s...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Computational mathematics and modeling Jg. 32; H. 4; S. 438 - 452
Hauptverfasser: Minin, Iu. B., Matveev, S. A., Fedorov, M. V., Zacharov, I. E., Rykovanov, S. G.
Format: Journal Article
Sprache:Englisch
Veröffentlicht: New York Springer US 01.10.2021
Springer Nature B.V
Schlagworte:
Applications of Mathematics
Benchmarks
Computational Mathematics and Numerical Analysis
Electric fields
Green's functions
Hankel matrices
Helmholtz equations
Integral equations
Linear systems
Mathematical analysis
Mathematical Modeling and Industrial Mathematics
Mathematics
Mathematics and Statistics
Optimization
Photonics
Solvers
Sparse matrices
Topology optimization
Python module
C++
performance
GMRES
Helmholtz equation
Toeplitz-like matrix
topology optimization
parallelization
GPU
ISSN:1046-283X, 1573-837X
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Generalized Minimal Residual Method (GMRES) was benchmarked on many types of GPUs for solving linear systems based on dense and sparse matrices. However, there are still no GMRES implementation benchmarks on Tesla V100 compared to GTX 1080 Ti ones or even for Toeplitz-like matrices. The introduced software consists of a Python module and a C++ library which enable to manage streams for concurrent computations of separated linear systems on a GPU (and GPUs). The GMRES solver is parallelized for running on a NVIDIA GPGPU accelerator. The parallelization efficiency is explored when GMRES is applied to solve (Helmholtz equation) linear systems based on the use of Green’s Function Integral Equation Method (GFIEM) for computing electric field distribution in the design domain. The proposed implementation shew the maximal speedup of 55 ( t ¯ = 0.017 s ) and of 125 ( t ¯ = 0.77 s ) for 1024 × 1024 (on GTX 1080 Ti) and 8192 × 8192 (on Tesla V100) dense Toeplitz matrices generated from GFIEM. 1024 × 1024 resolution provides accuracy 6.1% that can be acceptable according to testing and demonstrating on gradient computations and topology optimization. We open up possibilities for robust topology optimization of passive photonic integrated components. That has the advantage, e. g., of faster and more accurate designing photonic components on a PC without a supercomputer.
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:1046-283X
1573-837X
DOI:10.1007/s10598-022-09545-2