PNNL

Abstract

Quantum chemistry by its nature was a compute intensive field of science from the very beginning. It was therefore also one of the early adopters of parallel computing. After over twenty years of parallel computing in the field the challenges are changing in a number of areas including the disappearance of single core/processor machines, the rise of hybrid computing, the pressure from "green" computing, and machines with levels of concurrency that were unimaginable just ten years ago. This raises the questions like what have learned from parallel computing so far, and is there a way to address the next levels of parallelism? In this chapter key parallel algorithms of molecular quantum chemistry will be considered, such as the replicated and distributed data Fock builders, matrix-matrix multiplication, matrix diagonalization, and integral transformation routines. They will be used to illustrate valuable lessons about what can be expected from parallel computing. Finally, parallel infrastructures based on directed acyclic graphs will be considered that currently attract considerable interest as a future way to manage parallelism. Recent developments in that area will be discussed and their suitability for future parallel quantum chemistry computing demands considered.