Skip to content
#graph neural networks Book Open access

AtomFlow: Accelerating GNN-Based Molecular Dynamics with ab initio Accuracy on Fugaku Supercomputer

Sep 2026 · International Conference on Parallel Processing · 0 citations · 24 references
Computer Science

TL;DR

This paper proposes AtomFlow, a holistic acceleration for a representative GNN-based MD model on the Fugaku supercomputer, and redesigns the inference workflow with a framework-free implementation to avoid the overhead of the PyTorch framework on Fugaku.

Abstract

Molecular dynamics (MD) simulations with ab initio accuracy have become indispensable for studying microscopic phenomena. Although graph neural network (GNN)-based MD models provide higher accuracy than deep neural network (DNN)-based models, their complex message-passing architectures introduce substantial computational and communication bottlenecks for large-scale inference. In this paper, we propose AtomFlow, a holistic acceleration for a representative GNN-based MD model on the Fugaku supercomputer. Our key optimizations include: (1) we design a fused domain-specific EC kernel that avoids materializing large edge-related intermediate tensors; (2) we design an inter-process communication operator that leverages Fugaku’s uTofu interconnect to accelerate atomic-feature exchange; and (3) we redesign the inference workflow with a framework-free implementation to avoid the overhead of the PyTorch framework on Fugaku. Experimental results show that AtomFlow achieves up to a 9.77 × speedup over the baseline implementation, reaching up to 26.14 nanoseconds per day for a 576,000-atom system with 12,000 nodes on Fugaku.

Read PDF

Similar papers

Preprint Oct 2026

Wisevolve Potential Engine: Machine-Learning Molecular Dynamics at 100 Million Atom-Steps per Second on a Single GPU

The Wisevolve Potential Engine (WPE) is presented, which uses information from GPUMD to select and reuse a GPU execution policy tailored to the model and simulation stage, and support model- and stage-specific specialization as a practical way to accelerate GPU potential evaluation in general-purpose molecular dynamics...

He-Kai Bu, Zhe-Yong Fan · 0 citations
Preprint Oct 2026

Train for Accuracy, Execute at Scale: Architecture-Preserving Inference for Equivariant Atomistic Foundation Models

Equivariant atomistic foundation models provide broadly transferable interatomic potentials trained against quantum-mechanical reference data, but their repeated execution at simulation scale remains computationally and memory intensive. We present Symmetrix-XL, an inference engine that scales pretrained MACE checkpoin...

Lei Fu, Zi-Hui Feng, Yong-Heng Li et al. · 2 citations
#machine learning Preprint Oct 2026

Learning consistent molecular mechanics force fields from first principles

Classical force fields (FFs) remain the workhorse for large-scale simulations even as machine-learned interatomic potentials (MLIPs) approach ab initio accuracy. They decompose total configuration energies into simple effective interactions whose parameters are traditionally assigned based on atom or bond types, enabli...

Berkay Gunes, Leif Seute, Jigyasa Nigam et al. · 0 citations
Open access Sep 2026

Local Variational Graph Coarsening for Machine Learning Coarse-Grained Molecular Dynamics Simulations.

This work introduces a multilevel graph-coarsening framework for coarse graining, inspired by recent advances in graph reduction with spectral and cut guarantees, and provides a data-driven bottom-up computational framework for the development of systematically improvable CG potentials with controlled accuracy and phys...

Soumya Mondal, Subhanu Halder, Debarchan Basu et al. · 0 citations
#machine learning Preprint Oct 2026

FlashCart: Fast Cartesian Tensor Products for Equivariant Interatomic Potentials

Machine-learned interatomic potentials extend atomistic simulations beyond the length- and timescales accessible to electronic-structure methods. However, the computational cost of equivariant architectures limits the local correlations they can represent in practice and therefore their achievable accuracy. Here we int...

V. Zaverkin, P. Goodarzi, S. Sukhomlinov et al. · 0 citations
#software testing Preprint Sep 2026

Agent-E2MD: Autonomous Translation of Interatomic Potential Equations into Physically Validated Pair Styles for Molecular Dynamics in LAMMPS

Agent-E2MD is introduced, a knowledge-guided agentic workflow that translates user-specified interatomic models into executable LAMMPS pair styles and demonstrates that the software architecture for an interatomic potential can be inferred from its physical and mathematical structure.

Bilvin Varughese, Orcun Yildiz, Aditya Koneru et al. · 0 citations

Related blog posts

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.