Cray Computer Corporation

~ the obituary ~

Cray was a word before it was a machine, at least anywhere outside the few dozen institutions that owned one. It meant the fastest computer that existed, the way a brand name sometimes gets borrowed to mean a whole category, and that is how it arrived in a Turkish magazine or a school textbook. The hardware itself stayed in weapons labs and weather bureaus on other continents. This grave is about the company the man started after the company with his name on it had gotten too big to let him work.

Seymour Cray left Control Data in 1972 to found Cray Research, then gave up the chief executive job in 1980 so he could go back to designing. The Cray-3 was his next machine after the Cray-2 and it was meant to be about twelve times faster, somewhere near sixteen gigaflops, at a 500 MHz clock. Work started in Chippewa Falls and moved to Colorado Springs in 1988 to keep it away from the other project competing for the same budget. That arrangement lasted about a year before Cray Research decided it could not carry both.

On 15 November 1989 Cray Research issued shares in a new company, Cray Computer Corporation, to its own stockholders, retained ten percent and put an eighty-five million dollar note behind the development work. Seymour Cray went with the lab and ran the design. He was not a shareholder in the company that carried his name for the second time, and worked for it under contract.

The Cray-3 was finished. One was built, delivered to the National Center for Atmospheric Research in Boulder on 24 May 1993 with four processors and 128 megawords of memory, and it went there as a loan rather than a sale because no customer had ordered one. Nobody ever did. The company filed a voluntary Chapter 11 petition in the District of Colorado on 24 March 1995, after a private placement of up to twenty-five million dollars failed to come together and it stopped having liquid assets. All 340 people at the Colorado Springs site were let go.

~ the case file ~

~ gallium arsenide ~

Seymour Cray's whole career had been built on a conservative rule, which was that the exotic part of a machine should be the architecture and the packaging, not the parts you order. He used proven logic and won by arranging it better than anyone else. For the Cray-3 he broke his own rule and chose gallium arsenide, because silicon could not switch fast enough to get him his factor of twelve and GaAs could.

The problem is that gallium arsenide in 1989 had no industry behind it. Yields were immature and there was no second source for a company of a few hundred people to fall back on when the parts came back wrong. The machine also had to be physically tiny for the signals to arrive on time, which made assembly a matter of stacking modules into a volume smaller than a domestic appliance and then cooling the result. Every one of those decisions was correct in isolation. Together they meant that the schedule belonged to a semiconductor process nobody else was using.

~ one machine, on loan ~

NCAR did not pay for the Cray-3 and never planned to. The unit sat in Boulder as a demonstration and a debugging platform, with static RAM that misbehaved and processors that would not stay stable, which is a fair description of a prototype and a poor one of a product. A supercomputer company at that stage survives on a launch customer who wants the machine badly enough to live through the first year of it. Cray Computer never found one, and the machine at NCAR was the closest it came.

~ the cold war ended ~

The customers for a sixteen gigaflops machine in 1989 were national laboratories and the defense agencies attached to them, and most of that money existed because of a standoff that stopped in 1991. The Warsaw Pact dissolved. The budgets that had bought every previous Cray were suddenly arguments that had to be made rather than line items that renewed. The Cray-4 was already in design at a one gigahertz target and it was aimed at a market that was contracting while it was being built.

What finished the company was ordinary. It needed twenty-five million dollars from institutional investors to keep the Cray-4 going and the placement did not close. A hardware company with a payroll and no cash has about a week of options. The bankruptcy is reported as having consumed around half a billion dollars over its life and more than four hundred jobs, for one delivered machine that was never sold.

~ after the filing ~

Seymour Cray started again, founding SRC Computers in 1996 to build from commodity microprocessors, which after a career of avoiding them reads like an acceptance of what had actually happened to the field. He did not get to work on it long. On 5 October 1996 he died of injuries from a car accident on the road near Colorado Springs, about a week after his seventy-first birthday. Cray Research had by then been bought by Silicon Graphics, and the name went on without either of them.

@}-->--

              )
             (_)
            .-'-.
            |   |
            |   |
            |   |
            |   |
          __|   |__   .-.
       .-'  |   |  `-:   :
      :     `---'     :-'
       `-._       _.-'
           '""""""

« previous  ·  random  ·  next »