Analysis of Sparse Matrix-Vector Multiplication Using Iterative Method in CUDA

Abstract

Scaling up the sparse matrix-vector multiplication has been at the heart of numerous studies in both academia and industry. The massive parallelism of graphics processing units offers tremendous performance in many highperformance computing applications. In this work, we discuss performance analysis for parallel implementation of sparse matrix-vector multiplication using the conjugate gradient algorithm that are efficiently implemented on the NVIDIA CUDA architecture to exploit the massive compute power of today's GPUs. The results show that in comparison to the parallel CPU implementations, the parallel version of the conjugate gradient algorithm on GPU is in average 30 times faster depending on computational kernels.

Fast Iterative Solvers for Large Compressed-Sparse Row Linear Systems on Graphics Processing Unit

“…These linear algebraic operations implemented in CUDA C++ by the authors are denoted in the following by Alinea BLAS(1) for vector-vector operations (Saxpy, Dot and Norm) and Alinea BLAS (2) for SpMV operations. Reference [32] gives an analysis of SpMV using iterative method for real number arithmetic. Numerical methods, including domain decomposition methods (DDM), require the solution of linear systems.…”