Benutzer: Gast  Login
Titel:

Simulation-Tuned Time and Energy Optimisation for Multi-Site and 3-body Molecular Dynamics Simulations with AutoPas

Dokumenttyp:
Konferenzbeitrag
Art des Konferenzbeitrags:
Vortrag / Präsentation
Autor(en):
Newcome, Samuel James
Abstract:
Within the field of Molecular Dynamics (MD) exist numerous algorithms, such as Linked Cells and Verlet Lists, as well as parallelisation strategies -- with none performing optimally for every simulation. This caveat has motivated the development of the node-level auto-tuning and algorithm-selection library AutoPas, which aims to automatically select the algorithmic configuration with the optimal time-to-solution or energy efficiency [Gratl et al., 2022]. In addition, AutoPas can change the algorithmic configuration over the course of a simulation and in different regions of space. In this work, we will present AutoPas itself, as well as our research into the effective application of AutoPas to multi-site MD, where molecular models consist of small rigid bodies [Newcome et al., 2023], and 3-body MD, where force kernels are computed using triplets of molecules [Axilrod & Teller, 1943]. Within multi-site MD, we will highlight the challenges which arise from designing high-performing data structures for these complex multi-site molecules as well as the vectorisation of relevant force kernels. Within 3-body MD, we will present our novel Linked Cells shared-memory parallelisation schemes.
Stichworte:
AutoPas; Molecular Dynamics; Multi-Site Molecular Dynamics; 3-Body Molecular Dynamics
Kongress- / Buchtitel:
CECAM Flagship Workshop: Perspectives and challenges of future HPC installations for atomistic and molecular simulations
Datum der Konferenz:
21.02.2024
Jahr:
2024
Quartal:
1. Quartal
Jahr / Monat:
2024-02
Monat:
Feb
Reviewed:
nein
Sprache:
en
TUM Einrichtung:
Department of Computer Science
 BibTeX