H196-0004
An efficient GPU solver for highly heterogeneous flows

Wednesday, 16 December 2020
Poster
Lucas Bessone, UDELAR, CENUR Litoral Norte, Departamento de Matemática y Estadística del Litoral, Salto, Uruguay; UDELAR, CENUR Litoral Norte, Departamento del Agua (Water Department), Salto, Uruguay, Pablo Gamazo, Universidad de la República, CENUR Litoral Norte, Departamento del Agua, Salto, Uruguay, Marco Dentz, IDAEA-CSIC, Barcelona, Spain, Mario Storti Dr., Centro de Investigación de Métodos Computacionales (CIMEC), CONICET-UNL, Santa Fe, Argentina, Pablo Ezzatti, Universidad de la República, Uruguay, Facultad de Ingeniería, Montevideo, Uruguay and Julián Ramos, Universidad de la República de Uruguay, CENUR Litoral Norte, Departamento del Agua, Salto, Uruguay
Abstract:
In this work, we present a simple, fully GPU-based implementation for solving an elliptical problem with highly jumping coefficients. The solver consists of a very simple and efficient implementation of the Cell Centered Multi Grid (CCMG) method used as a multi-level preconditioner in the Conjugate Gradient (PCG) method. The construction is carried out in the Finite Volume Method context on uniform Cartesian meshes. Unlike pure CCMG, this combination is more robust for these types of problems. Comparing with the PCG with the Semicorasening Multi Grid (SMG) preconditioner provided by Hypre interface in parallel CPU with 36 cores, we found a speedup of 7X for problems of size over 50 MCells using a single GPU Nvidia Tesla V100.
<< Previous Abstract | Next Abstract >>