Efficient molecular dynamics simulations with many-body potentials on graphics processing units (1610.03343v2)

Abstract: Graphics processing units have been extensively used to accelerate classical molecular dynamics simulations. However, there is much less progress on the acceleration of force evaluations for many-body potentials compared to pairwise ones. In the conventional force evaluation algorithm for many-body potentials, the force, virial stress, and heat current for a given atom are accumulated within different loops, which could result in write conflict between different threads in a CUDA kernel. In this work, we provide a new force evaluation algorithm, which is based on an explicit pairwise force expression for many-body potentials derived recently [Phys. Rev. B 92 (2015) 094301]. In our algorithm, the force, virial stress, and heat current for a given atom can be accumulated within a single thread and is free of write conflicts. We discuss the formulations and algorithms and evaluate their performance. A new open-source code, GPUMD, is developed based on the proposed formulations. For the Tersoff many-body potential, the double precision performance of GPUMD using a Tesla K40 card is equivalent to that of the LAMMPS (Large-scale Atomic/Molecular Massively Parallel Simulator) molecular dynamics code running with about 100 CPU cores (Intel Xeon CPU X5670 @ 2.93 GHz).

• The paper introduces a novel algorithm for efficiently evaluating forces with many-body potentials on GPUs by accumulating contributions within a single thread to avoid write conflicts.
• The open-source GPUMD implementation demonstrates high performance, comparable to ~100 CPU cores for the Tersoff potential and simulating millions of atoms at speeds up to 25 nanoseconds per day.
• This work reduces computational barriers for large-scale simulations of complex materials, enabling more detailed studies and showing promise for adaptation to other potentials and future GPU advancements.

Efficient Molecular Dynamics Simulations with Many-Body Potentials on Graphics Processing Units

Recent advances in computational physics have enabled the acceleration of molecular dynamics (MD) simulations using graphics processing units (GPUs), primarily focusing on pairwise potential evaluations. The paper "Efficient molecular dynamics simulations with many-body potentials on graphics processing units" addresses the less-tackled challenge of optimizing force evaluations for many-body potentials. This is crucial because many-body interactions, epitomized by potentials such as the Tersoff and Stillinger-Weber, are fundamental to accurately modeling the properties of materials at the atomic level.

Overview of the Study

The authors propose a novel algorithm for evaluating forces in the context of many-body potentials on GPUs. The conventional approach to force evaluation involves multiple loops for each atom's interactions, leading to potential write conflicts when implemented in parallel computational environments like CUDA. By leveraging an explicit pairwise force expression recently derived in prior work, the proposed method accumulates interaction contributions within a single thread, guaranteeing the absence of write conflicts.

The paper introduces "GPUMD," an open-source code implementation based on these formulations. For the Tersoff potential, the performance benchmarking on a Tesla K40 GPU demonstrated computational efficiency comparable to running LAMMPS on approximately 100 CPU cores, highlighting the effectiveness of the GPU acceleration.

Performance and Implications

The proposed method exhibits strong performance with impressive scaling characteristics. For large atomic systems, the single-precision variant of the code achieved computational speeds that far surpass traditional CPU-based implementations, reaching up to 25 nanoseconds per day for systems with millions of atoms. Such advancements significantly reduce the computational barriers to simulating large, complex material systems—a necessary step for accurate materials design and discovery.

The strong numerical results affirm the implemented algorithms' efficiency, providing a pathway for extensive simulations that were previously deemed computationally prohibitive. By improving the performance of simulations involving many-body interactions, this work paves the way for more comprehensive studies in materials science, particularly involving phenomena such as defect formation, stress distribution, and thermal transport properties.

Future Directions

Given the adaptability of the proposed force evaluation strategy to other many-body potentials beyond Tersoff and Stillinger-Weber, future developments could extend to even more complex interactions, such as those described by the REBO potential. As GPU architectures evolve, the methods introduced here could be further optimized to leverage new hardware capabilities, possibly incorporating advanced techniques such as synchronous threading and memory management.

Moreover, real-world applications, including the simulation of nanomaterials, polymers, and biomolecules, stand to benefit significantly from these efficiencies. The paper hints at the vast potential for future research in integrating these algorithms into multi-scale modeling frameworks, thus enhancing the predictive power of simulations across diverse material systems.

In summary, the paper's contributions represent a significant methodical evolution in the simulation of many-body interactions using GPUs. While practical applications currently center around materials science, the implications hold promise for any field requiring accurate and efficient simulation of complex systems at the atomic level.