Very-Large-Scale Density Functional Theory on Massively Parallel GPUs
Inkoo Kim
Computational Science and Engineering Team, Innovation Center, Samsung Electronics
Very-Large-Scale Density Functional Theory on Massively Parallel GPUs
Inkoo Kim
Computational Science and Engineering Team, Innovation Center, Samsung Electronics
Modern graphics processing units (GPUs) provide an unprecedented level of computing power, enabling significant advancements in computational chemistry. In this talk, I will present a high-performance multi-GPU implementation of the Kohn–Sham density functional theory (DFT) and time-dependent DFT (TDDFT), along with their analytical nuclear gradients. The discussion will focus on algorithms optimized for efficient Fock matrix construction on massively parallel systems, leveraging multiple parallel models (MPI, OpenMP and CUDA) to achieve optimal scalability with increasing material size, considerably reducing computational time. To illustrate the effectiveness of this approach, I will present a benchmark TDDFT study on a biological protein consisting of 4,353 atoms with 40,518 Gaussian-type atomic orbitals, performed at the wB97X/def2-SVP level of theory. This study demonstrates favorable parallel efficiencies, with >70% efficiency on up to 64 GPUs and approximately 30% with 256 GPUs, using a high-speed distributed system equipped with 256 Nvidia A100 GPUs. These results highlight the potential of modern high-performance computing systems to tackle very-large-scale quantum chemical simulations efficiently.