GPU Accelerated Quantum Molecular Dynamics Simulation for Efficient Electron-Force-Field Model
Zhanbo Li, Yuhao Zhang, M. Li
Traditional molecular dynamics simulations have been providing crucial support to the fields of physics, chemistry, and materials science for many years. However, recent research efforts have increasingly focused on microphysical phenomena and quantum effects. Given the cost and complexity of experimental studies, quantum molecular dynamics simulations have emerged as a viable alternative for research. Despite their potential, existing simulation projects are hindered by the limited performance of CPUs. To overcome this limitation, we present an advanced and efficient implementation of Electron Field Force (EFF) model. The Electron Field Force (EFF) model, based on wave-particle duality and numerical approximation, can effectively provide an accurate correction to the conventional results derived from traditional methods. In this paper, a GPU-accelerated version of the EFF model is proposed to enhance simulation capabilities significantly. Our implementation has achieved a remarkable speedup of 90x on a single A100 GPU node. Additionally, we have developed a scalable multi-GPU version using MPI+CUDA technology for further scalability, which greatly extends the scope of research in quantum molecular dynamics.
International Conference on Artificial Intelligence and Computer Information Technology (AICIT)•Sep 20, 2024•0 citations