GPU-Accelerated Preconditioned Iterative Linear Solvers
Department of Computer Science & Engineering, University of Minnesota, Minneapolis, MN 55455, USA
Minnesota Supercomputer Institute, University of Minnesota, Report umsi-2010-112, 2010
@techreport{li2010gpu,
title={Gpu-accelerated preconditioned iterative linear solvers},
author={Li, R. and Saad, Y.},
year={2010},
institution={Technical report, University of Minnesota}
}
This work is an overview of our preliminary experience in developing high-performance iterative linear solver accelerated by GPU co-processors. Our goal is to illustrate the advantages and difficulties encountered when deploying GPU technology to perform sparse linear algebra computations. Techniques for speeding up sparse matrix-vector product (SpMV) kernels and finding suitable preconditioning methods are discussed. Our experiments with an NVIDIA TESLA C1060 show that for unstructured matrices SpMV kernels can be up to 12 times faster on the GPU than on the host Intel Xeon E5504 Processor. Overall performance of the GPUaccelerated Incomplete Cholesky (IC) factorization preconditioned CG method can outperform its CPU counterpart by a much smaller factor, up to 3, and GPU-accelerated Incomplete LU (ILU) factorization preconditioned GMRES method can achieve a speedup nearing 4. However, with better suited preconditioning techniques for GPUs, this performance can be significantly improved.
December 19, 2011 by hgpu
