GPU-Acceleration of the WEST Code for Simulating Electronic Excitations in Large, Heterogeneous Materials
ORAL
Abstract
We present a massively parallel, GPU-accelerated implementation of many-body perturbation theory and time-dependent density-functional theory (TDDFT) in the WEST code (https://west-code.org) for the simulation of the excited states of large molecules and materials. Outstanding performance and scalability are achieved by employing a hierarchical parallelization strategy, overlapping communications with computations, and utilizing the near-sightedness principle in selected portions of the code. We show the results of large-scale calculations using full-frequency G0W0, the Bethe-Salpeter equation, quantum defect embedding theory, and TDDFT with hybrid functionals and analytical forces, for systems containing over a thousand atoms. Finally, we delve into our experience of switching GPU programming models to attain enhanced performance portability targeting the first exascale supercomputers deployed in the United States.
* This work is supported by MICCoM, as part of the Computational Materials Sciences Program funded by the U.S. Department of Energy, Office of Science, Basic Energy Sciences.
–
Publication: 1. Journal of Chemical Theory and Computation 18 (2022): 4690--4707
2. Modelling and Simulation in Materials Science and Engineering 31 (2023): 063301
3. arXiv:2309.03513 [cond-mat.mtrl-sci]
Presenters
-
Victor Yu
Argonne National Laboratory
Authors
-
Victor Yu
Argonne National Laboratory
-
Yu Jin
University of Chicago
-
Giulia Galli
University of Chicago
-
Marco Govoni
Argonne National Laboratory, University of Modena and Reggio Emilia