ePubs

The open archive for STFC research publications

Full Record Details

DOI 10.5286/raltr.2008023
Persistent URL http://purl.org/net/epubs/work/49086
Record Status Checked
Record Id 49086
Title A fast and robust mixed precision solver for the solution of sparse symmetric linear systems
Contributors
Abstract The main bottleneck for emerging computing architectures is memory bandwidth. The amount of data moved around within a sparse direct solver can be approximately halved by using single precision arithmetic. However, the cost of this is a potential loss of accuracy in the solution of the linear systems. Double precision iterative methods preconditioned by a single precision factorization can enable the recovery of high precision solutions more quickly than a sparse direct solver run using double precision arithmetic. The gains from the reduced memory bandwidth are expected to be particularly prominent on multicore machines where the ratio of computational power to bandwidth is higher. In this paper, we develop a practical algorithm to apply such a mixed precision approach and suggest parameters and techniques to minimise the number of solves required by the iterative recovery process. These experiments provide the basis for our new code HSL_MA79 - a fast, robust, mixed precision sparse symmetric solver that will be included in the mathematical software library HSL. Numerical results for a wide range of problems from practical applications are presented.
Organisation CSE , CSE-NAG , STFC
Keywords Fortran 95 , FGMRES , multifrontal method , iterative refinement , sparse symmetric linear systems , mixed precision , Gaussian elimination
Funding Information
Related Research Object(s):
Licence Information:
Language English (EN)
Type Details URI(s) Local file(s) Year
Report RAL Technical Reports RAL-TR-2008-023. STFC, 2008. MixedPrecFastRobust.pdf 2008