Massive Parallelism: the Hardware for Computational Chemistry? [chapter]

M. F. Guest, P. Sherwood, J. A. Nichols
2001 Astrophysics and Space Science Library  
Efficient and portable "shared memory" interface -Scalable -Adheres to standards or predicted standards -Insulates chemistry programmer from CS intricacies -Placed in the public domain to encourage adoption These tools, developed at PNNL over the past few years, are the basis for the success of the NWChem project and the wealth of functionality coming on line. ftp.pnl.gov/pub/permanent MOAO Pµν guess orbitals ga_dgemm Integrals V XC V Coul V 1 e PeIGS If Converged SCF Module Fρσ Daresbury
more » ... tory COST-Chemistry D3 Workshop, Perugia. June 1997 Department for Computation and Information Time on 512 Delta nodes: 156 secs Daresbury Laboratory COST-Chemistry D3 Workshop, Perugia. June 1997 Department for Computation and Information GAMESS-UK is the general purpose ab initio molecular electronic structure program for performing SCF-and MCSCF-gradient calculations, together with a variety of techniques for post Hartree Fock calculations. The program is derived from the original GAMESS code, obtained from Michel Dupuis in 1981 (then at the NRCC), and has been extensively modified and enhanced over the past decade. Energies and geometries of medium to large ground-state molecules at Hartree-Fock Level: -Integrals recomputed or held in core rather than stored on disk -Memory (scales as n 2, n=no. of basis functions), leads to limitations on workstations (approx. 2000 basis functions on 128 Mbyte machines) -Cost of Fock Matrix build scales as n 4 (small systems) to ca n 2.5 asymptotically. Cost of matrix operations within SCF sheme n 3 (matrix diagonalisation, MMO etc.) -Integral generation highly parallel (DLB); O(n 3 ) operations potential bottleneck on MPP machines -O(n) techniques for integral evaluation; matrix OPs the bottleneck.
doi:10.1007/978-94-010-0864-8_17 fatcat:odl427uklfgtzpkq5yqbeyckh4