High performance computational chemistry: NWChem and fully distributed parallel applications

Abstract

We describe the development, implementation and performance of the NWChem computational chemistry package. Targeting both present and future generations of massively parallel processors (MPP), this software features a variety of efficient parallel algorithms for a broad range of methods commonly used in computational chemistry. The emphasis throughout is on scalability and the distribution, as opposed to the replication, of key data structures. To facilitate such capabilities, we describe a shared non-uniform access memory model which simplifies parallel programming while at the same time providing for portability across both distributed- and shared-memory machines. As an example of the capabilities of NWChem, we outline the development and performance of a highly efficient and scalable algorithm for conducting selfconsistent field Density Functional calculations on molecular systems. A performance analysis of this module in calculations on the zeolite fragment Si28O67H30 (1673 basis functions) is presented for representative MPP systems, the IBM-SP2, Kendall Square Research KSR-2, the Intel Paragon and Cray T3D. Finally we consider, based on these figures, the scale of performance likely to be achieved on the next generation of MPP systems.