Supercomputer

The invention: 

A computer that had the greatest computational power that then existed.

The person behind the invention: 

Seymour R. Cray (1928-1996), American computer architect and designer




The Need for Computing Power

Although modern computers have roots in concepts first proposed
in the early nineteenth century, it was only around 1950 that they became
practical. Early computers enabled their users to calculate equations
quickly and precisely, but it soon became clear that even more
powerful computers—machines capable of receiving, computing, and
sending out data with great precision and at the highest speeds—
would enable researchers to use computer “models,” which are programs
that simulate the conditions of complex experiments.
Few computer manufacturers gave much thought to building the
fastest machine possible, because such an undertaking is expensive
and because the business use of computers rarely demands the
greatest processing power. The first company to build computers
specifically to meet scientific and governmental research needs was
Control Data Corporation (CDC). The company had been founded
in 1957 by William Norris, and its young vice president for engineering
was the highly respected computer engineer Seymour R.
Cray. When CDC decided to limit high-performance computer design,
Cray struck out on his own, starting Cray Research in 1972. His
goal was to design the most powerful computer possible. To that
end, he needed to choose the principles by which his machine
would operate; that is, he needed to determine its architecture.


The Fastest Computer

All computers rely upon certain basic elements to process data.
Chief among these elements are the central processing unit, or CPU
(which handles data), memory (where data are stored temporarily
before and after processing), and the bus (the interconnection between
memory and the processor, and the means by which data are
transmitted to or from other devices, such as a disk drive or a monitor).
The structure of early computers was based on ideas developed
by the mathematician John von Neumann, who, in the 1940’s,
conceived a computer architecture in which the CPU controls all
events in a sequence: It fetches data frommemory, performs calculations
on those data, and then stores the results in memory. Because it
functions in sequential fashion, the speed of this “scalar processor”
is limited by the rate at which the processor is able to complete each
cycle of tasks.
Before Cray produced his first supercomputer, other designers
tried different approaches. One alternative was to link a vector processor
to a scalar unit. Avector processor achieves its speed by performing
computations on a large series of numbers (called a vector)
at one time rather than in sequential fashion, though specialized
and complex programs were necessary to make use of this feature.
In fact, vector processing computers spent most of their time operating
as traditional scalar processors and were not always efficient at
switching back and forth between the two processing types.
Another option chosen by Cray’s competitors was the notion of
“pipelining” the processor’s tasks. A scalar processor often must
wait while data are retrieved or stored in memory. Pipelining techniques
allow the processor to make use of idle time for calculations
in other parts of the program being run, thus increasing the effective
speed. A variation on this technique is “parallel processing,” in
which multiple processors are linked. If each of, for example, eight
central processors is given a portion of a computing task to perform,
the task will be completed more quickly than the traditional von
Neumann architecture, with its single processor, would allow.
Ever the pragmatist, however, Cray decided to employ proved
technology rather than use advanced techniques in his first supercomputer,
the Cray 1, which was introduced in 1976. Although the
Cray 1 did incorporate vector processing, Cray used a simple form
of vector calculation that made the technique practical and easy to
use. Most striking about this computer was its shape, which was far
more modern than its internal design. The Cray 1 was shaped like a
cylinder with a small section missing and a hollow center, with
what appeared to be a bench surrounding it. The shape of the machine
was designed to minimize the length of the interconnecting
wires that ran between circuit boards to allow electricity to move the
shortest possible distance. The bench concealed an important part
of the cooling system that kept the system at an appropriate operating
temperature.
The measurements that describe the performance of supercomputers
are called MIPS (millions of instructions per second) for scalar
processors and megaflops (millions of floating-point operations per
second) for vector processors. (Floating-point numbers are numbers
expressed in scientific notation; for example, 1027.) Whereas the fastest
computer before the Cray 1 was capable of some 35 MIPS, the
Cray 1 was capable of 80 MIPS. Moreover, the Cray 1 was theoretically
capable of vector processing at the rate of 160 megaflops, a rate
unheard of at the time.


Consequences

Seymour Cray first estimated that there would be few buyers for
a machine as advanced as the Cray 1, but his estimate turned out to
be incorrect. There were many scientists who wanted to perform
computer modeling (in which scientific ideas are expressed in such
a way that computer-based experiments can be conducted) and
who needed raw processing power.
When dealing with natural phenomena such as the weather or
geological structures, or in rocket design, researchers need to make
calculations involving large amounts of data. Before computers,
advanced experimental modeling was simply not possible, since
even the modest calculations for the development of atomic energy,
for example, consumed days and weeks of scientists’ time.
With the advent of supercomputers, however, large-scale computation
of vast amounts of information became possible. Weather
researchers can design a detailed program that allows them to analyze
complex and seemingly unpredictable weather events such
as hurricanes; geologists searching for oil fields can gather data
about successful finds to help identify new ones; and spacecraft
designers can “describe” in computer terms experimental ideas
that are too costly or too dangerous to carry out. As supercomputer
performance evolves, there is little doubt that scientists will
make ever greater use of its power.


Seymour R. Cray

Seymour R. Cray was born in 1928 in Chippewa Falls, Wisconsin.
The son of a civil engineer, he became interested in radio
and electronics as a boy. After graduating from high school in
1943, he joined the U.S. Army, was posted to Europe in an infantry
communications platoon, and fought in the Battle of the
Bulge. Back from the war, he pursued his interest in electronics
in college while majoring in mathematics at the University of
Minnesota. Upon graduation in 1950, he took a job at Engineering
Research Associates. It was there that he first learned
about computers. In fact, he helped design the first digital computer,
UNIVAC.
Cray co-founded Control Data Corporation in 1957. Based
on his ideas, the company built large-scale, high-speed computers.
In 1972 he founded his own company, Cray Research Incorporated,
with the intention of employing new processing methods
and simplifying architecture and software to build the
world’s fastest computers. He succeeded, and the series of computers
that the company marketed made possible computer
modeling as a central part of scientific research in areas as diverse
as meteorology, oil exploration, and nuclear weapons design.
Through the 1970’s and 1980’s Cray Research was at the
forefront of supercomputer technology, which became one of
the symbols of American technological leadership.
In 1989 Cray left Cray Research to form still another company,
Cray Computer Corporation. He planned to build the
next generation supercomputer, the Cray 5, but advances in microprocessor
technology undercut the demand for supercomputers.
Cray Computer entered bankruptcy in 1995.Ayear later
he died from injuries sustained in an automobile accident near
Colorado Springs, Colorado.

See also :  Apple II computer; BINAC computer; Colossus computer;
ENIAC computer; IBM Model 1401 computer; Personal computer; Seymour R. Cray



Further Reading

Slater, Robert. Portraits in Silicon. Cambridge, Mass.: MIT Press,
1987.




Understanding Supercomputing