In this paper we discuss the general question of the portability of Molecular Dynamics codes for diffusive systems (liquids) on parallel computers of the APE family. The intrinsic single precision arithmetics of the today available APE platforms does not seem to affect the numerical accuracy of the simulations, while the absence of integer addressing from the CPU to individual nodes puts strong constraints on the possible programming strategies. As a test case, we report the results of the simulation of the dynamics of 512 molecules of liquid butane (C4H10) ta room temperature. After 30 to 50 ps of equilibration, the system was followed along four long trajectories, each one lasting more than 1.3 ns. The effective CPU time corresponding to the simulation of a trajectory of 1 ns on the Torre (512 nodes ∼ 25 Gigaflops) was ∼ 50 hours. The CPU time can be substantially reduced (by almost a factor 3), if the APE-assembler micro-code of the most time-consuming part of the program is carefully optimized. © 1997 Elsevier Science B.V.
La Penna, G., Minicozzi, V., Morante, S., Rossi, G., Salina, G. (1997). Molecular dynamics with the massively parallel APE computers. COMPUTER PHYSICS COMMUNICATIONS, 106, 53-68 [10.1016/S0010-4655(97)00086-6].
Molecular dynamics with the massively parallel APE computers
MINICOZZI, VELIA;MORANTE, SILVIA;
1997-01-01
Abstract
In this paper we discuss the general question of the portability of Molecular Dynamics codes for diffusive systems (liquids) on parallel computers of the APE family. The intrinsic single precision arithmetics of the today available APE platforms does not seem to affect the numerical accuracy of the simulations, while the absence of integer addressing from the CPU to individual nodes puts strong constraints on the possible programming strategies. As a test case, we report the results of the simulation of the dynamics of 512 molecules of liquid butane (C4H10) ta room temperature. After 30 to 50 ps of equilibration, the system was followed along four long trajectories, each one lasting more than 1.3 ns. The effective CPU time corresponding to the simulation of a trajectory of 1 ns on the Torre (512 nodes ∼ 25 Gigaflops) was ∼ 50 hours. The CPU time can be substantially reduced (by almost a factor 3), if the APE-assembler micro-code of the most time-consuming part of the program is carefully optimized. © 1997 Elsevier Science B.V.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.