Software and hardware required to obtain optimal performance for simulation research are being developed. These include the development of software for massively parallel Multiple Instruction/Multiple Data (MIMD) machines, microcoding commercial processors to obtain optimal performance, and the development of workstation cluster hardware and associated software for optimal performance as a parallel computer. Massively parallel high performance computers hold great promise for the future of high speed scientific computing. An Intel i860-based machine of this type is being used for algorithm development and scientific computing. For a system with 128 processors, simulation speedup of a factor of 75 (60% efficiency) is the current level of performance for a macromolecular system with 14,000 atoms and a 13.5 Angstrom nonbonded interaction cutoff distance. Methods for this class of machine are being further designed and tested, with the goal of using this system for future research. Workstation clusters provide a highly competitive environment in terms of cost performance for macromolecular simulations. A workstation cluster based on the Hewlett-Packard-730 machine has been assembled. Parallel software is being developed and is being evaluated as a function of network connectivity (Ethernet, Token ring, or fiber ring (FDDI). The software under development for the parallel cluster includes both CHARMM for macromolecular simulation, and LIGHT, an NIH-developed ray-trace raster image molecular graphics program.
