Our work “Accelerating large-scale excited-state GW calculations on leadership HPC systems” (authored by M. Del Ben, C. Yang, Z. Li, F. H. da Jornada, S. G. Louie, and J. Deslippe) has been selected as a finalist for the 2020 Gordon Bell Prize. In this work, the GPU-accelerated BerkeleyGW has been successfully running at full scale of the Summit machine at Oak Ridge Leadership Computing Facility, utilizing 27,648 GPUs, and reaching 105.9 petaFLOP/s in double precision, 52.7 of the machine peak. A silicon divacancy structure of over 2,700 atoms and over 10,000 electrons is solved with a time-to-solution of 10 minutes.
This work has been presented at the virtual SC20 meeting. The video presentation is available at: https://youtu.be/lMOg-oFVCWg