We have studied portability, efficiency and accuracy of a standard Molecular Dynamics simulation on the SIMD parallel computer APE100. Computing speed performance and physical system size range have been analyzed and compared with those of a conventional computer. Short range and long range potentials have been considered, and the comparative advantage of different simulation approaches has been assessed. For long range potentials, APE turns out to be faster than a conventional computer; large systems can be conveniently simulated using either the cloning approach (up to ~ 10^5 particles) or a domain decomposition with the systolic method. In the case of short range potentials and systems with diffusion (like a liquid), APE is convenient only when using a large number of processors. In a special case (a crystal without diffusion), a specific domain decomposition technique with frames makes APE advantageous for intermediate and large systems. Using the latter technique we have studied in detail the effect of different numerical error sources, and compared the accuracy of APE with that of a conventional computer.
Molecular Dynamics on APE100 / Barone, Luciano Maria; Riccardo, Simonazzi; Tenenbaum, Alexander. - In: COMPUTER PHYSICS COMMUNICATIONS. - ISSN 0010-4655. - STAMPA. - 90:(1995), pp. 44-58. [10.1016/0010-4655(95)00076-R]
Molecular Dynamics on APE100
BARONE, Luciano Maria;TENENBAUM, Alexander
1995
Abstract
We have studied portability, efficiency and accuracy of a standard Molecular Dynamics simulation on the SIMD parallel computer APE100. Computing speed performance and physical system size range have been analyzed and compared with those of a conventional computer. Short range and long range potentials have been considered, and the comparative advantage of different simulation approaches has been assessed. For long range potentials, APE turns out to be faster than a conventional computer; large systems can be conveniently simulated using either the cloning approach (up to ~ 10^5 particles) or a domain decomposition with the systolic method. In the case of short range potentials and systems with diffusion (like a liquid), APE is convenient only when using a large number of processors. In a special case (a crystal without diffusion), a specific domain decomposition technique with frames makes APE advantageous for intermediate and large systems. Using the latter technique we have studied in detail the effect of different numerical error sources, and compared the accuracy of APE with that of a conventional computer.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.