Why a supercomputer is so `super'. Huge memory, blazing speed are two reasons
Cupertino, Calif. — Unlike the fictional Superman, supercomputers are real-life heroes who win real-life praise the old-fashioned way: They earn it. These machines, which play a major role in research at NASA Ames Research Center in Mountain View, Calif., and at Lawrence Livermore National Laboratory in Livermore, Calif., and soon will run at close to the speed of light, can:
Simulate the disastrous leak in space shuttle Challenger's solid-fuel rocket booster, and deliver a computer-screen image with hundreds of times the resolution of mission control's tracking film.
Simulate the birth of a star, the collision of two galaxies, and the makeup of Earth's atmosphere. (Using supercomputer models, for example, scientists can estimate what effect freon and other pollutants have on the atmosphere.)
Simulate human vision so engineers can build robots that could replace astronauts in the repair of space stations.
Help forecast global weather patterns based on data from millions of locations.
Simulate the shape of oil basins, enabling oil companies to map oil and gas reserves.
What makes supercomputers ``super''? For one thing, they can perform hundreds of millions of calculations per second. For another, supercomputers have such large memories that they can instantly report the results of these calculations to a high-resolution display screen.
The more information a computer can handle at once, the clearer the computer-screen image that the information yields. Such images might include a space shuttle, a jet, or an automobile that is undergoing design changes on a computer screen. Shorter wires, faster calculations
The Cray-2 supercomputer, which experts say is the most powerful computer in the world, can complete up to 2 billion calculations per second (that's 50,000 times faster than a personal computer). Furthermore, it can store an amount of information equal to that contained in a branch of the San Jose Public Library.
The supercomputer is C-shaped instead of rectangular to shorten the length of internal wires. Each wire in its 15-mile network is precisely cut to reduce the time that a signal spends traveling to the different parts of the computer.
Its blazing speed and large memory make the Cray-2, which costs $17 million, an ideal tool for solving some of the most troublesome puzzles facing scientists. Cray Research Inc. of Minneapolis, Minn., introduced the Cray-2 last September. Other companies that make less powerful supercomputers include Control Data, NEC, Fujitsu, and Hitachi.
Victor Peterson, director of aerophysics at NASA Ames Research Center, says that scientists there have pioneered several uses of the high-speed machines.
``The first application that Ames programmed for the Cray-2 allowed us to redesign the space shuttle's main engine,'' he says. ``NASA wants to boost the thrust of the space shuttle so that eventually we can double its payload. The Cray-2 looked for the most efficient fuel path through the main engine.''
More recently, Ames programmed its Cray-2 to redesign the solid-fuel rocket boosters for future shuttle flights.
Another project at Ames is to better understand the element oxygen. ``The computer confirms what we already [el36l]know about oxygen and gives us a look at how oxygen reacts to other molecules,'' Peterson explains. ``We used the Cray-2 to determine that the `halo' that forms around the shuttle as it cruises the upper atmosphere probably resulted from its collision with stray oxygen molecules.'' Simulating Jupiter flyby
Mostly, however, the supercomputer helps Peterson's team study aerodynamics -- the motion of air and other gases and how they affect spacecraft.
``The Galileo spacecraft, for example, will send out a probe to explore Jupiter,'' he says. ``It is not possible to recreate the conditions of a probe flying through the atmosphere of Jupiter in our ground-based test facilities. So we simulate the Jupiter flyby with the Cray-2 computer.''
Researchers at Lawrence Livermore use supercomputers to simulate fusion, the process by which the nuclei of two atoms join to produce energy. A Cray-2 also has allowed scientists there to mimic an exploding star and the oscillation of a uranium nucleus that has been struck by a proton, one of the elementary particles in the nuclei of atoms.
Lawrence Livermore has harnessed big computers of one sort or another since the 1950s, says Hans Bruijnes, deputy director of the lab's National Magnetic Fusion Energy Computer Center.
He says the supercomputer of that day -- the IBM 701 -- was used to model nuclear explosions mathematically. The device filled a large machine room, stored and processed data with vacuum tubes, and ``had the speed and memory capacity of an Osborne portable computer.''
The Cray-2 has given Livermore researchers a way to test nuclear weapons without detonating them, according to George Michael, a computer scientist at the lab. Elsewhere, Michael notes, scientists will use a special purpose supercomputer called the GF10 to predict the theoretical mass of the proton.
``Supercomputers will make revolutionary changes in every field of technology,'' Michael predicts.
Seymour Cray designed the Cray-1 -- introduced in 1976 -- in his head rather than relying on the help from another computer, the common approach in computer design. Given the complexity of the Cray-1, this was a superhuman task.
His most recent invention, the Cray-2, affirms a trend in computer design that dates back to the early days of electronic computing: the shorter the distance between two circuits, the faster the computing speed. Cray packed 230,000 integrated circuits (or chips) into his super machine. Personal computers, on the average, contain about 150 chips.
These logic and memory chips, make up 320 logic and memory blocks in the Cray-2. Each block has eight layers of circuit boards stacked on top of one another.
In an instant, information-rich electronic signals can travel between chips along the length or width of the 1-by-4-by-8 inch block. This part of the information exchange is two-dimensional -- the limit of conventional computers.
But signals in the Cray-2 can also move vertically to a circuit board above or below. In conventional computers, the information must first travel to the edge of the circuit board, then up or down to the next board, then across the second board to reach its destination. It is the ability to operate in three dimensions -- plus such dense packing of chips -- that makes the Cray-2 so fast.
However, such dense packing generates heat -- enough, in fact, to melt the computer if it isn't cooled properly. Industrial-grade Fluorinert cools the Cray-2 to around 90 degrees F., or 60 degrees cooler than most computers. Some 150 gallons of the colorless, inert coolant are used. The Cray-2 gulps 150,000 watts of power and creates as much heat as 1,000 tightly grouped, glowing 150-watt light bulbs.
Cooling the chips is essential because the failure rate of semiconductors increases at higher temperatures.
Scientists say that the heart of the next generation of Cray supercomputer -- the Cray-3 -- will measure about one cubic foot, compared with about 20 cubic feet in the Cray-2. At peak performance, the new Cray model may be able to process information at close to the speed of light, the fastest that an electronic signal can travel.
The Cray-3, which scientists expect will be available in 1988, will probably use gallium arsenide transistors instead of silicon transistors. Gallium arsenide is more brittle than silicon and a more difficult material to make into chips, but it also allows transistors to switch on and off five to 10 times faster than the silicon chip.