A multi-GPU parallel computing framework for large-scale phase-field fracture simulations

doi:10.1016/j.cma.2026.118856

A multi-GPU parallel computing framework for large-scale phase-field fracture simulations

2026-02-20

0

PRE

AI

Daniel Shigueo Morikawa

DOI：10.1016/j.cma.2026.118856

Original

Original request for help

Save

Abstract

En 中文

• A multi-GPU framework is proposed for large-scale phase-field fracture with strong scalability and compact memory use. • An HCSR-SoA data structure enabling efficient data transfer and linear memory scaling is developed. • A unified RAM-VRAM scaling model is developed to predict memory demand and GPUs for given problem sizes. • Strong scaling on 8 GPUs achieves a 6.75× speedup and roughly 900× over a serial CPU. • The capacity of our framework for engineering-scale problems is demonstrated via a 300-million-DOF case.

Keywords:

multi-GPU computing

phase-field fracture

strong scaling

memory efficiency

large-scale simulations

A multi-GPU parallel computing framework for large-scale phase-field fracture simulations

Abstract

Journal

Organization