Since 1978
Published in Sarov (Arzamas-16), Nizhegorodskaya oblast

RUSSIAN FEDERAL
NUCLEAR CENTER -
ALL-RUSSIAN RESEARCH INSTITUTE
OF EXPERIMENTAL PHYSICS
 
 Русский |  English
ABOUT EDITORIAL BOARD PUBLICATION ETHICS RULES FOR AUTHORS AUTHORS ARCHIVE MOST RECENT ISSUE IN NEXT ISSUE PAPER OF THE YEAR




USE OF GRAPHIC ARITHMETIC ACCELERATORS FOR MOLECULAR DYNAMICS COMPUTATIONS USING MD SOFTWARE

B. L. Voronin, A. M. Erofeev, S. V. Kopkin, I. A. Kryuchkov, A. S. Rybkin, S. A. Stepanenko, V. V. Yuzhakov
VANT. Ser.: Mat. Mod. Fiz. Proc 2009. Вып.2. С. 55-61.

Applicability of computers with hybrid architecture to solve molecular dynamics problems is studied. Such computers have multiple-purpose processors and graphic arithmetic accelerators NVIDIA 8800GTX.
The paper analyzes the original algorithm and explains the way of dividing a code into fragments and measuring the runtimes of the code, as a whole and its separate fragments using multiple-purpose processors and arithmetic accelerators.
The paper shows that it is possible to use computers with hybrid architecture to solve MD problems. The MD software adaptation has been performed.
A hybrid computer allows reduction of the computational process duration (Morse potential) by a factor of 4.10, 2.67, and 1.49, as compared to one core, two cores and four cores, respectively.
The computational process duration (Ackland potential) is reduced by a factor of 6.64, 3.26, and 1.98 owing to a hybrid computer, as compared to one core, two cores and four cores, respectively.

Keywords: graphic arithmetic accelerator, processor, program code, performance, molecular dynamics.

APPLICATION OF GRAPHIC ARITHMETIC ACCELERATORS IN METHODOLOGICAL MONTE-CARLO TRANSPORT SIMULATIONS

I. A. Kryuchkov, S. P. Ognev, A. S. Rybkin, S. A. Stepanenko, V. V. Yuzhakov
VANT. Ser.: Mat. Mod. Fiz. Proc 2009. Вып.2. С. 62-68.

The appropriateness of a hybrid-architecture computer for speeding up Monte-Carlo simulations of eigenvalue problems is considered. A methodological program for neutron multiplication time constant calculations using the relaxation method has been used. This program is close in its nature to energy release simulation programs.
The program was modified and used on a hybrid-architecture computer with a MIMD- (universal processor) and SIMD- (graphic arithmetic accelerator - GAA) component, which allowed performing test simulations with one and two GAA, as well as with universal processor and GAA used in parallel.
These data demonstrate good prospects of MIMD and SIMD components (not only NVIDIA GeForce 8800GTX, but also AMD FireStream 9170, IBM PowerXCellSi (eDP) and other hybrid processors being developed) as applied to Monte-Carlo transport simulations.

Keywords: graphic arithmetic accelerator, GPU, processor, transport equation, Monte Carlo method.








[ Back ]


 
 
 
© FSUE "RFNC-VNIIEF", 2000-2024