Gilles Poncelet
PhD student
Ir. at UCLouvain in 2021

Main project: Towards (pre-)exascale computing: GPU acceleration in a massively parallel multiresolution framework for the incompressible Navier-Stokes equations
Funding: UCLouvain Assistant
Supervisor(s): Philippe Chatelain

Very large calculations on supercomputers have increasingly occupied a central role in scientific discoveries and engineering innovations. More and more machines are relying on hardware accelerators such as GPUs to further boost performances. However, as opposed to compute-bound problems (such as AI, ML,...), memory-bound problems (such as the resolution of the incompressible Navier-Stokes equations) are suffering increasingly important communication bottlenecks because of the latency inherent to memory transfers between the CPU and the GPU as well as the network bandwidth between the computer nodes. For such problems, simple hardware upgrades do not suffice and one needs to make changes to the algorithms and memory management strategies directly.

This project in particular will focus on the simulation of incompressible flows and will build upon a currently-being-developed multi-resolution open-source framework, murphy, and bring it to exascale-level scalability and computational intensity by relying on a unique combination of algorithmic and software tools: a wavelet-based block-structured multiresolution approach to exploit most out of the GPU acceleration in the upcoming pre-exascale facilities, such as LUMI.

IMMC main research direction(s):
Computational science
Fluid mechanics

eulerian-lagrangian method
numerical simulation, experimental models
particle methods
vortex method
wake flows

Research group(s): TFL