PARALLEL VERSION OF MNDO94 AVAILABLE
ON PSC's CRAY T3D and T3E.
ON PSC's CRAY T3D and T3E.
Carlos Gonzalez, Scientific Specialist, Pittsburgh Supercomputing Center
Richard L. Graham, Parallel Applications Specialist, Cray Research, Inc.
A parallel implementation of the quantum chemistry package MNDO94 is now available on PSC's CRAY T3D and T3E systems. MNDO94, developed by Dr. Walter Thiel, Chemistry Department, University of Zurich, Switzerland, offers the most popular semi-empirical Hamiltonians with capabilities to perform geometry optimizations, compute vibrational frequencies, electronic properties and energies and energies on a geometry grid.
The efficient parallelization of SCF energies, gradients, geometry optimizers, population analysis, numerical computation of the second derivative matrix (Hessian), reaction paths, potential energy grids, and more, has been achieved making use of distributed data structure algorithms. CPU intensive matrix algebra operations are carried out calling a series of fast routines contained in the ScaLAPACK library routines. The current version of MNDO94 is compatible with CRAY's shared memory communication libraries (SHMEM), as well as the message passing paradigms "PVM", and "MPI".
Preliminary results indicate very good performance. For example, a geometry optimization of the fullerene C240 (960 basis functions) that requires 7 hours and 16 minutes on a single processor of an Alpha DEC 8400, takes just 28 minutes on 32 processors on the CRAY T3D and about 9 minutes on the same number of CRAY T3E processors. In general, a speedup ratio of 3 has been observed on MNDO94 computations carried out on the T3E when compared to the corresponding results on the T3D.
If you are interested in trying parallel MNDO94, contact Dr. Carlos Gonzalez at cgonzale@psc.edu. Startup grants on the T3D and/or friendly-user grants on the T3E are available for testers.