|
|
|
| Current Issue |
|
|
| Resources |
|
|
|
Massively Parallel Computing
| Article
# : |
18721 |
|
|
Section : |
NATURAL SCIENCE
|
| Issue
Date : |
8 / 1991 |
2,762 Words |
| Author
: |
Sudip Dosanjh, Carl Diegert, and William Camp
Sundip Dosanjh, Carl Diegert, and William Camp are with Sandia
National Laboratories in Albuquerque, New Mexico. Dosanjh is
supervisor of the parallel computational sciences division;
Diegert is a staff member in the parallel computer science
division; and Camp is manager of the mathematics and
computational science department. |
Most computer users are familiar with some sort of desktop personal computer, or now, perhaps, the laptop computer. Somewhere at the distant end of the spectrum of computational speed and power has been the vector supercomputer, the top of the line computer with thousands of times greater computing capabilities [see "Supercomputers," THE WORLD & I, November 1986, p.172].
Until recently, vector supercomputers were the fastest computers in existence. The first vector processor, manufactured by Cray Research, Inc., in Mendota Heights, Minnesota, just 15 years ago, made a quantum leap in computer technology. Vector computers gained about a factor of 10 in speed by performing certain independent operations simultaneously. In addition, the new supercomputers used smaller and faster circuitry than previous generations of computers. The microcircuitry of these computers was so dense and generated so much heat that the computers relied on extremely complicated cooling systems.
Vector supercomputers have been the tool used for pioneering new applications of compute modeling to a widely diverse assortment of scientific and technological areas, ranging from climate modeling to designing airplanes, helping to build more efficient combustion engines, and searching for oil. Recently they have been used by Hollywood to create special effects in movies. Further increase in supercomputer speeds will broaden their applications and impact.
However, vector processors are close to their peak performance. The clock speed on a computer is approximately the time it takes to perform a single instruction; this determines the performance capability of vector supercomputers. During the past decade, clock speeds on vector supercomputers have increased by only a factor of 2. While a factor of 2 is a major improvement in many fields, it is not in the computer industry. By contrast, the clock speed of microprocessors inside personal, desktop computers has improved by factors of 10. The slow increase of supercomputer speeds is attributable to a fundamental physical barrier--namely, the speed of light. Clock speeds on vector supercomputers are already measured in nanoseconds. One nanosecond (0.000 000 001 seconds) is approximately the time it takes light to travel a foot (.3 meter). Dramatic increase in vector supercomputer performance will not occur without major technological breakthroughs (such as high-temperature superconductivity) that allow the size of logic circuitry to decrease significantly.
To overcome the barriers confronting vector processing,
...
Read Full Article
Look for this article in Ask.com
|
|
