Research

High Performance Computing

GPU Computing

GPU devices provides a great computing power with a low power consumption. Many applications may benefit from GPU acceleration.

• Use GPU as accelerators for compute intensive parts of applications
• Optimize code, algorithm and numerical methods to fit GPU architecture

Hybrid computing

The recent heterogeneous architectures provide both CPU and accelerators (GPU, co-processors) that must be exploited concurrently in numerical applications.

• Distribute computations on both CPU and co-processors (GPU)
• Efficient usage of full hybrid clusters
• 'In-situ' parallelism

Hybrid remeshed particle methods (semi-Lagrangian particle method)

High order remeshing formulas

• Analysis of high spatial order remeshing formulas with dimensional splitting
• Use GPUs to handle high numerical cost

Hybrid and multi-scale solver for passive scalar turbulent transport

• Use different scales for flow (coarse scale) and scalar transport (fine scale) computations
• Exploit different architectures for flow (CPUs) and transport (GPUs) solvers

Multi-scale solver for multiphase turbulent flows

• Multi-scale coupling between fluid flow and density transport

Collaborations

Unstationary CFD around aerodynamic profiles (ONERA)

Optimizing and up-scaling a research code, NextFlow, developped at ONERA. The aim of this code is to demonstrate feasibility of LES methods for simulating turbulent flows in realisstic aerodynamic configurations.

• Multi-GPU code based on Finite Volumes methods with high order polynomial reconstruction of conservatives variables.
• Asynchronous co-processing on uused CPU cores

NCI calculations using promolecular density (ICMR and ATOS)

Développment and optimization on GPU of a specific numerical method for studying ligand-protein interaction. The main objective is to integrate this code into a genetic algorithm for modecular docking.

Yales2 GPU porting (CORIA)

Preliminary study for GPU porting of this two-phase combustion DNS simulation code.