Header

UZH-Logo

Maintenance Infos

Towards electronic structure-based ab-initio molecular dynamics simulations with hundreds of millions of atoms


Schade, Robert; Kenter, Tobias; Elgabarty, Hossam; Lass, Michael; Schütt, Ole; Lazzaro, Alfio; Pabst, Hans; Mohr, Stephan; Hutter, Jürg; Kühne, Thomas D; Plessl, Christian (2022). Towards electronic structure-based ab-initio molecular dynamics simulations with hundreds of millions of atoms. Parallel Computing, 111:102920.

Abstract

We push the boundaries of electronic structure-based ab-initio molecular dynamics (AIMD) beyond 100 million atoms. This scale is otherwise barely reachable with classical force-field methods or novel neural network and machine learning potentials. We achieve this breakthrough by combining innovations in linear-scaling AIMD, efficient and approximate sparse linear algebra, low and mixed-precision floating-point computation on GPUs, and a compensation scheme for the errors introduced by numerical approximations. The core of our work is the non-orthogonalized local submatrix method (NOLSM), which scales very favorably to massively parallel computing systems and translates large sparse matrix operations into highly parallel, dense matrix operations that are ideally suited to hardware accelerators. We demonstrate that the NOLSM method, which is at the center point of each AIMD step, is able to achieve a sustained performance of 324 PFLOP/s in mixed FP16/FP32 precision corresponding to an efficiency of 67.7% when running on 1536 NVIDIA A100 GPUs.

Abstract

We push the boundaries of electronic structure-based ab-initio molecular dynamics (AIMD) beyond 100 million atoms. This scale is otherwise barely reachable with classical force-field methods or novel neural network and machine learning potentials. We achieve this breakthrough by combining innovations in linear-scaling AIMD, efficient and approximate sparse linear algebra, low and mixed-precision floating-point computation on GPUs, and a compensation scheme for the errors introduced by numerical approximations. The core of our work is the non-orthogonalized local submatrix method (NOLSM), which scales very favorably to massively parallel computing systems and translates large sparse matrix operations into highly parallel, dense matrix operations that are ideally suited to hardware accelerators. We demonstrate that the NOLSM method, which is at the center point of each AIMD step, is able to achieve a sustained performance of 324 PFLOP/s in mixed FP16/FP32 precision corresponding to an efficiency of 67.7% when running on 1536 NVIDIA A100 GPUs.

Statistics

Citations

Dimensions.ai Metrics
14 citations in Web of Science®
16 citations in Scopus®
Google Scholar™

Altmetrics

Downloads

32 downloads since deposited on 07 Oct 2022
14 downloads since 12 months
Detailed statistics

Additional indexing

Item Type:Journal Article, not_refereed, original work
Communities & Collections:07 Faculty of Science > Department of Chemistry
Dewey Decimal Classification:540 Chemistry
Scopus Subject Areas:Physical Sciences > Software
Physical Sciences > Theoretical Computer Science
Physical Sciences > Hardware and Architecture
Physical Sciences > Computer Networks and Communications
Physical Sciences > Computer Graphics and Computer-Aided Design
Physical Sciences > Artificial Intelligence
Uncontrolled Keywords:Artificial Intelligence, Computer Graphics and Computer-Aided Design, Computer Networks and Communications, Hardware and Architecture, Theoretical Computer Science, Software
Language:English
Date:1 July 2022
Deposited On:07 Oct 2022 10:48
Last Modified:28 May 2024 01:40
Publisher:Elsevier
ISSN:0167-8191
OA Status:Hybrid
Free access at:Publisher DOI. An embargo period may apply.
Publisher DOI:https://doi.org/10.1016/j.parco.2022.102920
Project Information:
  • : FunderEuropean Research Council
  • : Grant ID
  • : Project Title
  • Content: Published Version
  • Language: English
  • Licence: Creative Commons: Attribution 4.0 International (CC BY 4.0)