High-performance electron-dynamics simulations in GPU supercomputers with the INQ code

The study of materials under non-equilibrium conditions, such as those induced by intense laser pulses, requires real-time, time-dependent density functional theory (RT-TDDFT) simulations of unprecedented scale. We present INQ, a code designed from the ground up to harness the massive parallelism of modern GPU supercomputers for this purpose. INQ utilizes a plane-wave formalism to accurately describe the electronic excited states of atomic systems under strong external perturbations, achieving exceptional scalability and time-to-solution.

INQ’s current development effort is focused on pushing its scalability into the thousands of GPUs. A major challenge is the parallelization of the Fast Fourier Transforms (FFTs). The efficiency of FFTs is critical to the overall performance of RT-TDDFT in the plane-wave formalism. Furthermore, to move forward the accuracy of excited-state simulations, INQ is dedicated to incorporating hybrid functionals, that require non-local exchange integrals that also heavily relies on efficient FFTs.

In this talk, we detail our recent efforts to implement a novel, highly-optimized parallelization scheme for FFTs at the exascale level. We demonstrate how this advanced parallel FFT strategy, coupled with a robust implementation of hybrid functionals, enables INQ to simulate the excited-state dynamics of thousands of atoms on leading GPU architectures, establishing INQ as a premier tool for next-generation ab initio simulations of ultrafast phenomena.