The experience of using graphics processors for solving sparse systems of linear algebraic equations in the scope of hydrodynamic modeling of oil and gas fields
Keywords:
sparse algebra; graphics processors; iterative methods; multi? and many?core systems; parallel computing.Abstract
The impact of sparse matrices storage format on the execution time of basic operations of BiCGStab method with ILU(0) preconditioner on the NVIDIA Tesla K20X GPU is investigated. The approach to the parallelization of ILU(0) based on the combination of the block Jacobi method and level scheduling algorithm is introduced. Parallel solver for systems of linear algebraic equations implementing ILU(0) and CPR preconditioners is developed. Its effectiveness is evaluated on a hybrid computing system equipped with two NVIDIA Tesla K20X GPUs.Downloads
Published
2018-02-07
Issue
Section
INFORMATICS, COMPUTER ENGINEERING AND MANAGEMENT