Development of an Object-Oriented Program and a Special Purpose Parallel Computer for First Principles DVXa Calculations

Tohru SASAKIa, Umpei NAGASHIMAb* and Masaru TSUKADAc

aSony Kihara Reseach Center Inc.
1-10-4 Higashi-Gotannda, Shinagawa-ku, Tokyo 141-0022, Japan
Also at A-Priori Microsystems, Inc.
14-605, Keio Leading-edge Laboratory of Science and Technology,
3-14-1 Hiyoshi, Kohoku-ku, Yokohama, Kanagawa 223-8522, Japan
bComputational Science, National Institute for Advanced Interdisciplinary Research
1-1-4 Higashi, Tsukuba, Ibaraki 305-8562, Japan
cDepartment of Physics, Graduate School of Science, University of Tokyo
7-3-1 Hongo, Bunkyo-ku, Tokyo 113-8658, Japan
*e-mail:

(Received: April 6, 2000; Accepted for publication: June 20, 2000; Published on Web: October 30, 2000)

We have developed a program for first principles DVXa calculations using an object-oriented programming language C++. Based on the program we have also developed a special purpose parallel computer for DVXa calculations , which realizes super high performance and cost-down of large scale DVXa calculation.
The special purpose computer consists one host-board (Sparc Station10/74MHz) and 16 acceleration-boards. 4 TI320C40 processors are equipped on a board. The acceleration-board system has 64 processors connected to each other by VME-bus.
The matrix elements generation step where more than 90% calculation time spends was parallelized on the system. In the case of Si78B6H53cluster, the special purpose computer is almost 40 times faster than that by a lone Sparc Station10 processor. Speed up of the matrix elements generation step is more than 100 times though 64 times is the theoretical maximum. Thus, super-linear speed up has been observed.

Keywords: the First Principles DVXa Calculation, Special Purpose Computer, Object-Oriented Programming, C++, Super-linear Speed up


Abstract in Japanese

Text in Japanese

PDF file(186KB)


Author Index
Keyword Index
Return