A PCG Implementation of an Elliptic Kernel in an Ocean Global Circulation Model Based on GPU Libraries

TL;DR

This paper presents a GPU-accelerated implementation of an elliptic kernel for ocean circulation models, utilizing an inverse preconditioner to improve the efficiency of solving elliptic Laplace problems with the Conjugate Gradient method.

Contribution

It introduces a GPU-based solver for elliptic kernels in ocean models using an inverse preconditioner, enhancing computational performance and convergence.

Findings

GPU implementation improves solver speed

Convergence rate depends on grid resolution and coefficients

The method is easy to implement with scientific libraries

Abstract

In this paper an inverse preconditioner for the numerical solution of an elliptic Laplace prob- lem of a global circulation ocean model is presented. The inverse preconditiong technique is adopted in order to efficiently compute the numerical solution of the elliptic kernel by using the Conjugate Gradient (CG) method. We show how the performance and the rate of conver- gence of the solver are linked to the discretized grid resolution and to the Laplace coefficients of the oceanic model. Finally, we describe an easy-to-implement version of the solver on the Graphical Processing Units (GPUs) by means of scientific computing libraries and we discuss its performance.