Abstract

TEJA-JUAREZ, V. Leonardo  and  CRUZ, Luis M. de la. A Graphic Processing Unit (GPU) based implementation of an incompressible two-phase flow model in porous media. Geofís. Intl [online]. 2018, vol.57, n.3, pp.205-222. ISSN 2954-436X.

In this paper a parallelization strategy of a fully implicit simulator for the numerical solution of the incompressible two-phase flow model in porous media is presented using GPUs (Graphics Processing Units). The mathematical model is based on the mass conservation equations for the water and oil phases. Mathematical formulation of Pressure-Saturation is used to simplify the numerical model. The Finite Volume technique and the Newton-Raphson method are used to discretize and linearize the partial differential equations, respectively. The construction of the Jacobian directly on the GPU is proposed, which reduces the information that needs to be exchanged between the CPU (Central Processing Unit) and the GPU. The simulator uses libraries that already include methods that belong to the Krylov subspace to solve linear equations systems. The results of three benchmark problems by using different grid sizes are compared. The performance of the numerical code developed is also evaluated. Results of the GPU against the CPU indicate that the numerical simulator reached 22x of speed up to build the Jacobian, and 3x of speed up for executing the whole numerical code by using the GPU parallelization.

Keywords : Two-phase flow model; Newton-Raphson; Graphics Processing Units (GPU); Jacobian construction; speed up.

        · abstract in Spanish     · text in English     · English ( pdf )