Seminar by Timothée David-Cléris (CRAL)

Abstract

Numerical simulations are essential to our understanding of star and planet formation, which involve complex, multi-scale, non-equilibrium, and nonlinear multi-physical processes. Recently, the computational power of supercomputers has taken a quantum leap, reaching exascale—that is, one quintillion operations per second. In principle, this power makes it possible to resolve certain crucial questions regarding planet formation through simulations with unprecedented precision. To achieve this, it is necessary to develop a code based on algorithms capable of leveraging this new computational power.
The goal of this thesis is to develop Shamrock, the first exascale-targeted astrophysical code using multiple methods (particles or adaptive grids). The core of this work is the adaptation and optimization of a binary algorithm for searching for randomly distributed neighbors, which is fully parallelizable on architectures using graphics cards. In its current version, Shamrock achieves a parallel efficiency of over 90% for a Sedov test performed using the Smoothed Particle Hydrodynamics (SPH) method on 1,024 nodes, enabling the first simulations with 65 billion particles to be completed in 7 seconds per time step.
Further developments regarding the implementation of other physical processes, as well as tests of different schemes and their mathematical analysis, will also be presented.