04/30/90 THE GENIUS

Return to Past BW Coverage Table of Contents

THE GENIUS
Meet Seymour Cray, Father of the Supercomputer
There is a pad of graph paper on Seymour Cray's desk. Nothing else. No knick-knacks or family pictures. No trophies of personal achievement. Not even a calendar or pencil holder. Just a pad of graph paper on which he designs the fastest computers on earth.
That Spartan desk speaks volumes about Cray, a man of simple tastes and the concentration of a laser beam. These days, the 64-year-old ''Father of the Supercomputer'' is scrambling to finish his most far-reaching and riskiest creation, the Cray-3. At peak speeds, it should blast through a mind-bending 16 billion calculations a second -- eight times as fast as any computer like it. Yet all that power will be contained in an octagonal cabinet just three feet high and three feet across -- a Darth Vader-black marvel of muscular compaction. It will sell for as much as $30 million.
BIG TROUBLE. More than two years behind schedule and still months away from even the prototype stage, the Cray-3 project could be its maker's Waterloo. This time, the brilliant designer is betting on a radically new microchip technology, not just his renowned talent for eking maximum performance out of proven parts. The Cray-3 is expected to be the first commercially available computer to rely heavily on chips made from gallium arsenide (GaAs), not the silicon used in conventional semiconductors. Although much faster than silicon, GaAs is still unproven in full-scale computers and is devilish to work with.
The Cray-3 project has already become too risky for Cray Research Inc., the company Cray founded in 1976. Last fall, the Cray-3 and a team led by Seymour were spun off into Cray Computer Corp., capitalized with the assets of the project -- worth about $55 million -- and a loan of $100 million. But, racing for a late 1991 product debut, the startup will likely have to raise extra funding by next spring. Ominously, the prototype's due date has just slipped again, from summer to early next winter.
By the time it is ready for shipment, the Cray-3 and its brilliant inventor will face more competition than the supercomputer market has ever seen -- from Cray Research, which still has 60% of the market, from an IBM-backed startup, from Japan's NEC and Fujitsu, and from the many companies selling parallel-processing machines that deliver supercomputer power at a fraction of the price (page 85 43 ). ''Seymour Cray is in big trouble,'' asserts Jack Worlton, a supercomputing consultant and former scientist at Los Alamos National Laboratory. ''If his machine doesn't come out soon, it may be irrelevant.''
Irrelevant? Seymour Cray? It hardly seems likely. For 30 years, his machines and the ideas they embody have dominated high-performance computing. A shy and soft-spoken man, he is possessed with a single ambition: to build ever-faster computers that will attack ever-more-challenging problems. Cray's genius and singular drive have already marked indelibly the technology of this century. First used for cryptography and nuclear physics, supercomputers are quickly becoming indispensable to virtually every branch of science and engineering. Their ability to simulate all physical phenomena provides a new scientific method that joins theory and experimentation. Says Larry L. Smarr, the director of the National Center for Supercomputing Applications at the University of Illinois: ''I believe history will rank Cray with the likes of Thomas Edison.'' Like the fabled ''Wizard of Menlo Park,'' Seymour Roger Cray is an American original. Born in Chippewa Falls, a pleasant, hilly town in Wisconsin's northern dairy country, Seymour was the elder of two children. His father was the city engineer and his mother a homemaker. In most respects, Seymour had a typical Midwestern boyhood -- rambling through woods, bicycling down friendly streets, exchanging volleys of water bags with neighborhood buddies.
But Seymour was anything but typical. ''He had all this radio stuff in his basement,'' marvels boyhood chum Joseph C. Mandelert. ''None of us knew what it was.'' By age 10, using Erector Set parts and an old motor, he had built an automatic telegraph machine that could translate punched paper tape into Morse code signals.
Cray's parents nurtured his sharp, voracious mind. His father stocked a basement laboratory with chemistry sets and radio gear and permitted young Seymour to tinker into the wee hours. Eventually, a tangle of wires led from the lab to Seymour's second-floor bedroom, where his mother gave up trying to vacuum. His sister Carol, five years his junior, remembers what happened when anyone so much as touched the bedroom doorknob and set off an alarm: ''We'd hear his voice boom up from the basement, saying 'Who's entering my room?' '' When the Crays moved to another home, the wires got placed in a conduit behind the wall.
At Chippewa Falls High School, Cray didn't play sports and didn't chase girls. Science was his passion. When the physsics teacher was sick, Cray taught the class. Legend has it that he knew more about the subject than the teacher anyway. Of course, he graduated as the school's top science student, winning the Bausch & Lomb Science Award.
SLIDERULE ANGST. After a stretch in World War II -- first operating a radio in in Europe, then cracking Japanese codes in the Philippines -- Cray entered the University of Minnesota in Minneapolis. Fellow students recall his ability to slice through complex problems that left them scratching their heads. In recent years, Cray has taken pains to round out his personality, but in college he was, by most accounts, a 100% all-American nerd. In a recent -- and rare -- speech, Cray wryly recalled what he considered his ''social problem.'' He used a circular sliderule, and because it didn't hang from his belt in a leather case like the more popular straight ones, he fretted that he wouldn't be recognized on campus as a budding engineer.
In 1951, aged 26 and with a bachelor's degree in electrical engineering and a master's in applied mathematics, Cray made a fateful decision to go into computers. The commercialization of the ''electronic brain'' was just beginning as he joined Engineering Research Associates, a company started by William C. Norris, who later founded Control Data Corp. Operating from a converted glider factory in St. Paul, Norris and his team built computers for the Navy.
But when a series of mergers brought ERA under the purview of Sperry Rand Corp., developer of the Univac computer, Norris balked at the bureaucracy. In 1957, he set up Control Data in a rented corner of a warehouse across the river in Minneapolis. Norris had developed a knack for cultivating Cray's special talents and the young engineer soon followed. Like his boss, Cray found corporate bureaucracy an annoyance. Later, Norris recalls, he asked Cray to write a five-year plan for CDC. The result: ''Five year goal: Build the biggest computer in the world. One year goal: Achieve one-fifth of the above.''
At CDC, Cray created the 1604, one of the first computers to replace hot, bulky vacuum tubes with smaller, cooler transistors. Money was short, so the resourceful Cray shopped at a local electronics store to buy transistors for 37~ apiece. First shipped in 1960, the 1604 was snapped up by military and other research labs. It put CDC on the map.
BOMB-PROOF BRAINTANK. But Cray soon got fed up with CDC's own burgeoning bureaucracy. As the company grew, so did the number of managers, who always seemed to be in his way -- sometimes literally. ''People had to watch out for Seymour in the hall,'' says Robert E. Trier, a former CDC employee and now a consultant for the Research Consortium, a market researcher. ''He was so intent on a problem that he would run smack into people.'' In 1962, his prodigious talent in full bloom, Cray demanded a laboratory be built for him 100 miles east of Minneapolis, on a river bluff in his beloved Chippewa Falls. Norris, eager to accommodate his persnickety wizard, agreed.
More than red tape was bothering Cray. ''It was the time of the Cuban missile crisis,'' he later told a reporter. ''I wanted to get out of the big city because I might get my head blown off.'' The house he built in Chippewa Falls for his wife and children had a fallout shelter, a survival-size fuel tank, a Geiger counter, and an indoor swimming pool -- filled with potable water.
Norris' decision to let Cray and a small team hole up in the Wisconsin woods paid off handsomely. Cray turned out the CDC 6600, the world's first commercially available computer capable of executing 3 million program instructions per second -- far ahead of IBM's market-leading 7094 processor. The trick was the 6600's stripped-down internal design, which some experts view as the first use of reduced instruction-set computing (RISC). Now, more than 25 years later, RISC is being touted as a ''revolution'' in computing.
TV EYES. Cray excelled -- and still does -- at devising ways to pack computer circuits closely together. That enhances performance by cutting the time it takes electrical signals to travel between circuits. The catch is that circuits run hot, and the more densely they're packed, the harder it gets to cool them. Cray's ''real genius is clever packaging and heat removal,'' says Robert R. Borchers, head of computer operations at Lawrence Livermore National Laboratory, a longtime Cray customer. The 6600, for instance, was the first computer cooled with Freon, just like air conditioners. The 6600 looked cool, too. On its operator console, for instance, Cray replaced the traditional bank of switches and lights with a keyboard and a pair of round cathode-ray tubes -- big machine eyes. His next computer, the CDC 7600, featured doors of blue glass to show off the machine's innards. Cray was a fan of TV's Star Trek and his computers looked like they would fit in on the starship Enterprise. ''Everything Seymour does is esthetically beautiful,'' says Peter A. Gregory, a former top executive at Cray Research and now chief executive at Myrias Research Corp., the supercomputer maker.
But blazing performance, not aesthetics, was Cray's drawing card. The CDC 6600 inspired awe -- even from a begrudging Thomas J. Watson Jr., then chairman of IBM. He grew incensed when he saw puny CDC besting his legions of engineers and scientists. When the 6600 was introduced in August, 1963, Big Blue had just shoveled millions into developing a high-performance computer called Stretch, whose relatively poor performance had damaged IBM's reputation with leading scientific customers. In a blistering staff memo, Watson noted that the team that created the 6600 totaled ''only 34 people -- including the janitor . . . . Contrasting this modest effort with our own vast development activities, I fail to understand why we have. . . let someone else offer the world's most powerful computer.'' IBM finally spent millions trying to thwart Control Data. Its extreme tactics finally prompted CDC to file an antitrust suit. That was settled in CDC's favor, but not before it had encouraged the government and several aggrieved companies to sue IBM on antitrust grounds, too.
ALGEBRA ELVES. All Cray seemed to care about, though, was building fast computers. His children remember that on long auto trips he demanded total silence, apparently to think through technical problems. The young Crays, two girls and a boy, were even scolded for fiddling with the ashtray lids.
Cray's wife, Verene, worked hard to foster a sense of togetherness around Cray's obsessive work schedule. The daughter of a Chippewa Falls minister, she had married Cray right after his discharge from the Army. Whatever else was going on, she made sure that dinner at the Cray house was always a family affair, which her children remember fondly. But soon after dessert, Dad was back at the lab to work past midnight. His eldest child, Susan Cray Borman, recalls leaving him questions about her algebra homework on his desk in the evening, knowing she'd find answers waiting for her in the morning. ''It was like the elves had come,'' she says.
Cray's 7600 computer, announced in 1968 and capable of 15 million instructions per second, furthered Control Data's lead in scientific computing. But by 1972, Cray was getting restless again. That year Norris pulled the plug on Cray's planned 8600 computer in favor of another designer's project. ''I told Seymour he'd just have to wait until we got the other machine out,'' Norris says.
Cray had a better idea. With a handful of helpers, he set up Cray Research to design what would become the hugely successful Cray-1 supercomputer. Four years in the making, it would soon unseat CDC as the leader in scientific computing. And it remains a benchmark for all supercomputers.
The Cray-1 was like no other machine before it, inside or out. Viewed from above, it looked like a giant ''C,'' with an extended base that was covered in padded black vinyl and that doubled as a bench. Under the covers was the first commercially successful vector processor, which executed not one, but up to 32 arithmetic operations at a time. The result: a computer one-fourth the size of the CDC 7600 that could do 10 times the work. The young company earned back its original investment of $8.6 million in 1976 when the National Center for Atmospheric Research paid $8.8 million for the first Cray to be sold.
Even before his machine's smashing debut, however, the 51-year-old Cray had begun to do some soul-searching. ''I think a psychologist would say he had achieved a pretty high goal and he began looking around and seeing that there is just more to life,'' says his sister, Carol Cray Kersten. Cray mellowed, socializing more, going to parties, even hosting a few of his own. A friend says he'd invite guests from all walks of life, just to see them interact. ''I discovered people,'' he told a reporter.
Even so, Cray remained shy toward the press and by the late 1970s stopped giving interviews. A former colleague says that's when Cray began receiving threatening letters and phone calls condemning his machines' role in nuclear weapons research. Through a spokesman, Cray declined BUSINESS WEEK's requests for an interview, saying he's too busy working on the Cray-3 and doesn't want attention.
In 1975, Cray divorced Verene. A year later, he met Geri M. Harrand, now 54. An outgoing owner of a physical therapy business, Harrand helped Seymour develop new interests. They went to plays, windsurfed together, and skied with friends. In 1980, they married. ''He is still basically shy, but he's come out of his shell a little bit,'' she says.
Around this time, Cray also realized that he wasn't the right man to run a $60 million company that was growing 66% a year and drawing lots of attention from Wall Street. So in 1980, John A. Rollwagen took over as CEO. With engineering and business degrees, experience at CDC, and an easy, upfront manner, Rollwagen got along well with the finicky Cray.
BOOLEAN BOGGLE. Even as Cray Research's growth -- to $785 million by 1989 -- drew increasing attention, its founder became more and more mysterious. His refusal to talk to reporters gave rise to rumors that made him out to be the computer industry's answer to Howard Hughes -- an eccentric millionaire shunning humanity to pursue private obsessions.
Although many ''Seymour stories'' are based in fact, most are wildly exaggerated. One has Seymour burrowing mole-like through elaborate tunnel networks beneath his hillside home. In fact, he did dig a 12-foot reinforced hole in a lake bluff. It gave him some exercise and a place to store windsurfing gear. Another tale grew from reports that the inventor's mind is so fraught with the Boolean logic of computers that he can't slow down for the mundane details of everyday life. When shopping for a car, the story goes, he wanders into the showroom and simply picks the first car on his right. ''Ludicrous,'' say those who know him. Cray does prefer econoboxes to fancy cars, though, and according to his sister, mounts a watch on the dashboard rather than ordering a clock.
The most famous Cray legend concerns ceremonial bonfires. Each spring, it goes, Cray laboriously constructs a wooden sailboat from scratch. After a summer of sailing on Lake Wissota, he invites friends over for a lakeside party during which he torches the boat. The ritual is supposed to symbolize his need for constant renewal. Rollwagen has been largely responsible for spreading this tale but when pressed, he sheepishly admits that ''I made up the party.'' Daughter Carolyn Cray Bain says her father once did burn a boat that he'd built, but only because ''that was the easiest way to get rid of it.''
While the legends grew, Cray was doing the same old thing -- pushing computers to new levels of performance. In 1981, he let Rollwagen be chairman. Cray officially became a consultant to the company and devoted himself to the Cray-2, which then was intended to be the market's first GaAs computer. At the time, though, he couldn't get the GaAs chips he needed. So, he turned back to silicon for the Cray-2, which was finally completed in 1985. Still, he managed a tenfold performance jump over the Cray-1.
The troubled Cray-2 project was a turning point. Rollwagen began to wonder if the company's future didn't depend too much on Cray himself. He authorized a second, competing project, the Cray X-MP. Its chief designer, a Taiwanese computer whiz named Steve S. Chen, was told to squeeze more oomph out of the Cray-1 -- just in case the Cray-2 flopped. Chen's team succeeded in spades, completing a dual-processor system in 1982 that was almost three times as fast as the Cray-1. Seymour wasstunned. ''Somebody else made a machine faster than Seymour's,'' Rollwagen recalls. ''It had never happened.'' At first, Cray talked the X-MP down, insisting it wouldn't work. ''Then it did work, and he was very gracious,'' Rollwagen recalls. ''He said, 'I guess you guys really know how to do this.' ''
The Chen-Cray rivalry continued. Even before the Cray Y-MP -- the current world speed champ -- was finished, Chen in 1986 had begun working on a machine to rival Seymour's Cray-3. Chen ''got too big for his britches,'' says a former Cray Research executive. After spending more than $50 million on Chen's idea, Rollwagen killed the project in 1987 because its success depended on the unlikely event of major breakthroughs in five technologies. Chen quit and founded Supercomputer Systems Inc., which with backing from IBM, plans to ship its first machine by 1995.
Seymour Cray, meanwhile, forged ahead on the Cray-3. Things began looking up for GaAs technology in the mid-1980s when Gigabit Logic, a startup, began producing the first chips suitable for computers. ''The American system came to the rescue,'' a jubilant Cray told Cray Research shareholders in 1988. But GaAs was still not a sure thing. The material is extremely fragile, so few chips make it through the lengthy fabrication process. What's more, Seymour's design called for cramming 1,024 chips into a package measuring just 4 by 4 by 1/4 inches -- so delicate a job that special robots would be needed. A 16-processor Cray-3 would use 208 of those packages. The circuits would be so fast that no existing instruments could test them.
All this meant soaring costs -- just when Cray Research was struggling with soft orders and its first profit drop since going public in 1976. By 1989, the Cray-3 had eaten through $120 million and was still slipping its schedules. Again, Rollwagen had hedged his bets. Lester T. Davis, a Cray co-founder and close friend of Seymour's, was already overseeing a less-radical extension to the product line, the 16-processor Y-MP16. Although based on silicon, Davis says, it should run neck-and-neck with the GaAs Cray-3.
Rollwagen had little choice but to spin off the riskier project -- Seymour's. In return for assets associated with the Cray-3, Cray Research kept 10% of the new company. Seymour Cray evidently protested little and sold his Chippewa Falls house to settle permanently in Colorado Springs, where the project is based. And while he still takes time out for tennis several times a week, associates say he's more driven than at any time since the old days at Control Data. ''He seems to have found his role in life,'' says Neil Davenport, a Cray Research veteran who's playing the Rollwagen part at the new company. ''The rest of us are trying to support that as best we can.''
DEADLINE PRESSURE. Now, Cray Computer's race against time is getting critical. The company already is in danger of losing important customers. This fall, NEC Corp. plans to ship the SX-3, a silicon-based supercomputer that's claimed to go even faster than the Cray-3. There are rumors, too, that Fujitsu Ltd. has a GaAs super in the works. IBM's entry, via Chen, is looming, and many parallel-processing computers deliver low-cost supercomputing by combining thousands of microprocessors.
If Cray Computer slips too much, it may be unable to raise the hundreds of millions of dollars that will be needed to fund the Cray 4. That machine will use 64 GaAs processors, delivering 10 times the performance of the Cray-3 or the YMP-16. More important, the Cray-4 would prove conclusively that GaAs is a viable computer technology. Seymour Cray is staking his reputation on the belief that GaAs technology has the same potential to vastly improve computer performance that silicon chips had 25 years ago. Nobody's giving up on Seymour Cray. Although Wall Street has bid Cray Computer stock down to around 5 from its opening at 14 last summer, many analysts figure that Cray Computer can succeed -- if it can get the Cray-3 out the door next year. Then it would only need to sell 10 to 12 systems to recoup the initial investment and attract funds for the Cray 4. Analysts reckon that there are enough Cray zealots in government labs and industry to make that happen. ''These cult people will only buy from Seymour Cray,'' says Richard Sherman, head of Research Consortium.
And after a remarkable 30-year run, there are many true believers. Now in his mid-60s, Seymour shows no signs of slowing down. If his parents -- both active in their 90s -- are any guide, he has years of fruitful inventing ahead of him. But ''regardless of the outcome of the Cray-3, his track record will still be great,'' says old friend Les Davis. Seymour Cray already has guaranteed himself a place in the history books. From here on, it's all gravy.

Russell Mitchell in Minneapolis