Mixed-precision iterative refinement using tensor cores on GPUs to accelerate solution of linear systems

DOI10.1098/rspa.2020.0110zbMath1472.65174OpenAlexW3108141724WikidataQ104575678 ScholiaQ104575678MaRDI QIDQ5161159

A. Haidar, Nicholas J. Higham, Harun H. Bayraktar, Stanimire Z. Tomov, Jack J. Dongarra

Publication date: 29 October 2021

Published in: Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences (Search for Journal in Brave)

Full work available at URL: https://doi.org/10.1098/rspa.2020.0110

zbMATH Keywords

softwareGMREScomputational mathematicsLU factorizationGPU computingiterative refinementhalf precision arithmeticmixed precision solvers

Mathematics Subject Classification ID

Iterative numerical methods for linear systems (65F10) Numerical algorithms for specific classes of architectures (65Y10)

Related Items

Matrix Multiplication in Multiword Arithmetic: Error Analysis and Application to GPU Tensor Cores, Mixed precision algorithms in numerical linear algebra, Multiple-GPU accelerated high-order gas-kinetic scheme for direct numerical simulation of compressible turbulence, Five-Precision GMRES-Based Iterative Refinement, Acceleration of iterative refinement for singular value decomposition, Numerical stability of algorithms at extreme scale and low precisions, Random Matrices Generating Large Growth in LU Factorization with Pivoting, Exploiting Lower Precision Arithmetic in Solving Symmetric Positive Definite Linear Systems and Least Squares Problems, A Framework for Error-Bounded Approximate Computing, with an Application to Dot Products

Uses Software

• cuSOLVER

Cites Work

• Unnamed Item
• Unnamed Item
• Unnamed Item
• Unnamed Item
• Unnamed Item
• Accelerating scientific computations with mixed precision algorithms
• Towards dense linear algebra for hybrid GPU accelerated manycore systems
• Iterative refinement enhances the stability of \(QR\) factorization methods for solving linear equations
• Direct Methods for Sparse Matrices
• The university of Florida sparse matrix collection
• Mixed Precision Block Fused Multiply-Add: Error Analysis and Application to GPU Tensor Cores
• GMRES: A Generalized Minimal Residual Algorithm for Solving Nonsymmetric Linear Systems
• Iterative Refinement Implies Numerical Stability for Gaussian Elimination
• The Efficient Generation of Random Orthogonal Matrices with an Application to Condition Estimators
• Iterative refinement for linear systems and LAPACK
• Flexible Inner-Outer Krylov Subspace Methods
• Inexact Preconditioned Conjugate Gradient Method with Inner-Outer Iteration
• A New Analysis of Iterative Refinement and Its Application to Accurate Solution of Ill-Conditioned Sparse Linear Systems
• Accelerating the Solution of Linear Systems by Iterative Refinement in Three Precisions
• Solving Sparse Linear Systems with Sparse Backward Error
• Accuracy and Stability of Numerical Algorithms
• Random Matrices Generating Large Growth in LU Factorization with Pivoting
• Squeezing a Matrix into Half Precision, with an Application to Solving Linear Systems
• A Flexible Inner-Outer Preconditioned GMRES Algorithm
• Iterative Refinement in Floating Point
• Three-Precision GMRES-Based Iterative Refinement for Least Squares Problems
• Exploiting Lower Precision Arithmetic in Solving Symmetric Positive Definite Linear Systems and Least Squares Problems