|Name:||Role of HPC for the Development of an Innovative Industrial Application Based on the SPH Method|
|Time:||Wednesday, June 19, 2013
12:10 PM - 12:30 PM
|Room:||Hall 3 (Industry Track)
CCL - Congress Center Leipzig
|Speakers:||David Guibert, HydrOcean|
|Abstract:||The SPH-Flow code developed by Ecole Centrale de Nantes and HydrOcean relies on SPH methods to model complex free surface problems involving fast dynamics, such as the water entry of 3D objects of complex geometry. SPH methods suffer from high computational costs which increase dramatically in 3D engineering applications. To increase the use of SPH methods in industrial studies, large simulations must be solved in reasonable restitution time. Such simulations involve hundreds of millions of particles and, therefore, require important computational resources and an efficient parallelization.
Our code is based on a mesh-less method therefore classical parallel methods are not directly applicable. But parallel strategies could be introduced to take advantage of domain decomposition methods with ghost region as classically used in mesh-based method. With our new developments, simulations involving billion particles has been performed on tenth thousands cores. This work is being improved with hybrid OpenMP/MPI or GPU/MPI implementations.
This keynote will discuss how a SME used to launch computation on Tier2 machines could reach Tier0 eligibility and all the steps required to port an industrial application to Tier0 machines and to target manycores architectures including GPU cards.