Multigrid preconditioned asymmetric iterative solver for the incompressible SPH(2) for multi-GPU environments
Year: 2026
Type: article
Abstract: This paper presents the implementation of a multigrid-preconditioned asymmetric solver within the context of Incompressible Smoothed Particle Hydrodynamics (ISPH) method. The restriction operator maps particles to the background grid used for neighbor searching, followed by standard grid coarsening. Key contributions include the adjustment of the multigrid preconditioner to multi-GPU cluster environments, incorporating communication-computation overlap and communication-hiding techniques, and adapting a dynamic load balancer to respect the coarse-grid hierarchy for efficient domain decomposition using the slice-grid technique. The present work also shows the evaluation of the computational efficiency of the proposed method through dam-break simulations on three supercomputers - JAMSTEC’s Earth Simulator, Tongji University’s supercomputer, and the University of Tokyo’s Miyabi) - demonstrating weak-scaling efficiencies between 0.82 and 0.89. The multigrid preconditioner substantially improved solver performance, reducing the average number of iterations for solving the pressure from over 300 to nearly 60 across problem sizes ranging from 5 to 320 million particles, resulting in an overall speedup of approximately 2.3 times. Furthermore, we demonstrate the method’s capability for real-world engineering applications through the simulation of the 2011 tsunami inundation at the Fukushima Daiichi Nuclear Power Plant, involving 97 million particles on 32 GPUs. While the preconditioner proves highly effective in accelerating solver convergence, the study concludes by identifying limitations associated with coarse-grid constraints on load balancing and communication overhead, providing clear directions for future research.
Language: English
Subfield: Computational Mechanics
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $4,830
Awards: 「工学的方程式モデリング」の完全自動化への挑戦, G3361052475, Boosting emulator to estimate Millennium Typhoon damage in real time, Study on ensuring redundancy of bridges ccrossing rivers against lateral loads exceeding design conditions, Real-time prognosis of complex disasters through data science with embedded empirical knowledge of physics, G905527382
OpenAlex ID: W7130518237
