Will GPGPUs be Finally a Credible Solution for Industrial Reservoir Simulators?

Anciaux–Sedrakian, Ani; Eaton, Joe; Gratien, Jean-Marc; Guignon, Thomas; Havé, Pascal; Preux, Christophe; Ricois, O.

doi:10.2118/173223-ms

Cited by 7 publications

(4 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As far as the computing framework is concerned, we have implemented GMRES solver and the adaptive solve procedure on MCGSolver [20,21]. For exact and inexact LU factorizations, we used the library Eigen [22].…”

Section: Simulation Resultsmentioning

confidence: 99%

See 1 more Smart Citation

Adaptive linear solution process for single-phase Darcy flow

Anciaux–Sedrakian

Grigori

Jorti

et al. 2020

Oil Gas Sci. Technol. – Rev. IFP Energies nouvelles

View full text Add to dashboard Cite

This article presents an adaptive approach for solving linear systems arising from self-adjoint Partial Differential Equations (PDE) problems discretized by cell-centered finite volume method and stemming from single-phase flow simulations. This approach aims at reducing the algebraic error in targeted parts of the domain using a posteriori error estimates. Numerical results of a reservoir simulation example for heterogeneous porous media in two dimensions are discussed. Using the adaptive solve procedure, we obtain a significant gain in terms of the number of time steps and iterations compared to a standard solve.

show abstract

Section: Simulation Resultsmentioning

confidence: 99%

“…This is equivalent to the formulation obtained in the case of finite element methods, described in Section 3.3 of [6] (Formulas (19) and (20)). Due to the symmetry, the coupling terms are identical:…”

Section: Matrix Decomposition and Local Error Reductionmentioning

confidence: 99%

Adaptive linear solution process for single-phase Darcy flow

Anciaux–Sedrakian

Grigori

Jorti

et al. 2020

Oil Gas Sci. Technol. – Rev. IFP Energies nouvelles

View full text Add to dashboard Cite

show abstract

“…We have also implemented the corresponding modified versions, whereby one iteration of BiCGStab is performed before building the first 4s+1 bases vectors. These algorithms are developed in MCG Solver [18,19], a C++ based software package developed by IFPEN to provide linear solver algorithms for its industrial simulators in reservoir simulation, basin simulation or in engine combustion simulation. Table 3 shows the convergence results for the s-step BiCGStab versions on the matrices introduced previously with tolerance tol = 10 À8 except for the SPE10 matrix BiCGStab versions, which is equal to the s-step iterations times s. Note that in exact arithmetics, each s-step BiCGStab iteration is equivalent to s iterations of the classical BiCGStab method.…”

Section: Convergence Resultsmentioning

confidence: 99%

S-Step BiCGStab Algorithms for Geoscience Dynamic Simulations

Anciaux–Sedrakian

Grigori

Moufawad

et al. 2016

Oil Gas Sci. Technol. – Rev. IFP Energies nouvelles

View full text Add to dashboard Cite

-In basin and reservoir simulations, the most expensive and time consuming phase is solving systems of linear equations using Krylov subspace methods such as BiCGStab. For this reason, we explore the possibility of using communication avoiding Krylov subspace methods (s-step BiCGStab), that speedup of the convergence time on modern-day architectures, by restructuring the algorithms to reduce communication. We introduce some variants of s-step BiCGStab with better numerical stability for the targeted systems.Résumé -Méthodes s-step BiCGStab appliquées en Géosciences -Dans les simulateurs d'écoulement en milieu poreux, comme les simulateurs de réservoir et de bassin, la résolution de système linéaire constitue l'étape la plus consommatrice en temps de calcul et peut même représenter jusqu'à 80 % du temps de la simulation. Ceci montre que la performance de ces simulateurs dépend fortement de l'efficacité des solveurs linéaires. En même temps, les machines parallèles modernes disposent d'un grand nombre de processeurs et d'unités de calcul massivement parallèle. Dans cet article, nous proposons de nouveaux algorithmes BiCGStab, basés sur l'algorithme à moindre communication nommé s-step, permettant d'éviter un certain nombre de communication afin d'exploiter pleinement les architectures hautement parallèles.

show abstract

“…They reported an average of 20 minutes for the run time to solve a problem with 16 million of active blocks by using 4 GPUs. Anciaux-Sedrakian et al (2015) made a numerical study of different preconditioners such as: Polynomial, ILU and CPR-AMG which were implemented in a heterogeneous architecture (CPU-GPU). They emphasize two key points to obtain high performance in heterogeneous architectures; the first is to maximize the utilization and occupancy of the GPU and the second refers to minimize the high cost of transferring GPU data to the node with the CPUs and vice versa.…”

Section: Introductionmentioning

confidence: 99%

A Graphic Processing Unit (GPU) based implementation of an incompressible two-phase flow model in porous media

Teja-Juárez

Cruz

2018

GeofInt

View full text Add to dashboard Cite

En este trabajo se presenta una estrategia de paralelización de un simulador completamente implícito para la solución numérica del modelo de flujo bifásico incompresible en medios porosos usando unidades de procesamiento gráfico (GPU, por sus siglas en inglés). El modelo matemático está basado en las ecuaciones de conservación de masa para las fases agua y aceite. Se utiliza la formulación PresiónSaturación para simplificar el modelo numérico. La técnica de Volumen Finito y el método de Newton-Raphson se usan para discretizar y linealizar las ecuaciones diferenciales parciales, respectivamente. Se propone la construcción del Jacobiano directamente en la GPU, lo que reduce la información que debe intercambiarse entre la CPU (Unidad Central de Procesamiento CPU, por sus siglas en inglés) y la GPU. El simulador utiliza bibliotecas que ya incluyen los métodos del subespacio de Krylov para resolver sistemas de ecuaciones lineales. Se comparan los resultados de tres problemas de referencia utilizando diferentes tamaños de malla. También se evalúa el rendimiento del código numérico desarrollado. Los resultados de la GPU versus CPU indican que el simulador numérico alcanzó hasta 22x de aceleración para construir el Jacobiano y 3x de aceleración para ejecutar el código numérico completo usando la paralelización GPU.

show abstract

Will GPGPUs be Finally a Credible Solution for Industrial Reservoir Simulators?

Cited by 7 publications

References 13 publications

Adaptive linear solution process for single-phase Darcy flow

Adaptive linear solution process for single-phase Darcy flow

S-Step BiCGStab Algorithms for Geoscience Dynamic Simulations

A Graphic Processing Unit (GPU) based implementation of an incompressible two-phase flow model in porous media

Contact Info

Product

Resources

About