Toward an extreme-scale electronic structure system

Abstract

Electronic structure calculations have the potential to predict key matter transformations for applications of strategic technological importance, from drug discovery to material science and catalysis. However, a predictive physicochemical characterization of these processes often requires accurate quantum chemical modeling of complex molecular systems with hundreds to thousands of atoms. Due to the computationally demanding nature of electronic structure calculations and the complexity of modern high-performance computing hardware, quantum chemistry software has historically failed to operate at such large molecular scales with accuracy and speed that are useful in practice. In this paper, novel algorithms and software are presented that enable extreme-scale quantum chemistry capabilities with particular emphasis on exascale calculations. This includes the development and application of the multi-Graphics Processing Unit (GPU) library LibCChem 2.0 as part of the General Atomic and Molecular Electronic Structure System package and of the standalone Extreme-scale Electronic Structure System (EXESS), designed from the ground up for scaling on thousands of GPUs to perform high-performance accurate quantum chemistry calculations at unprecedented speed and molecular scales. Among various results, we report that the EXESS implementation enables Hartree–Fock/cc-pVDZ plus RI-MP2/cc-pVDZ/cc-pVDZ-RIFIT calculations on an ionic liquid system with 623 016 electrons and 146 592 atoms in less than 45 min using 27 600 GPUs on the Summit supercomputer with a 94.6% parallel efficiency.

High-Performance Multi-GPU Analytic RI-MP2 Energy Gradients

“…The present work builds on the existing RI-MP2 energy algorithm in EXESS detailed in a previous publication . The MO coefficients and eigenvalues are taken from the HF stage and a similar approach is used to calculate the B ia P and J PQ intermediates.…”

“…The EXESS software utilizes distributed MPI parallelism with one MPI process per GPU along with an additional coordinating process for dynamic work distribution . The majority of the intermediate values are stored in a MPI shared memory window to avoid duplication across ranks and reduce communication costs …”

“…In this Article, we present a novel high-performance algorithm and software implementation for the calculation of RI-MP2 energy gradients on multiple GPUs, which was integrated in the EXtreme-scale Electronic Structure System (EXESS) …”

“…A major paradigm shift in this hardware has occurred over the past decade with the widespread adoption of heterogeneous architectures. Accelerators such as graphical processing units (GPUs) now often provide the vast majority (>95%) of the computational power in the fastest machines. , Due to their fundamental architectural differences with CPUs, exploiting the computational capabilities of GPUs is challenging and requires a fundamental redesign of the complex algorithms underpinning quantum chemical methods …”

Benchmarking the MBE, FMO, and CPF GPU-Accelerated Fragmentation Methods for Accuracy and Parallel Time-to-Solution

“…To address this important question we have developed and implemented high performance, GPUaccelerated versions of the MBE and FMO methods in the EXtreme-scale Electronic Structure System (EXESS) program. 27,[30][31][32][33][34] Additionally, we have developed a new method, the Coulomb Perturbed Fragmentation (CPF) which, instead of iterating the ESP to self-consistency, calculates it only once using the densities of the individual fragments converged in vacuo, thus avoiding the outer SCF. All n-body contributions to the CPF energy are computed in the presence of the fixed ESP of the surrounding fragments.…”