Supercomputer Systems, Inc.

~ the goodbye ~

Steve Chen designed the Cray X-MP and then the Cray Y-MP, which between them were the machines most national labs meant when they said supercomputer in the 1980s. In September 1987 Cray Research cancelled the MP project he was running and he left. Four months later he had his own company in Eau Claire, Wisconsin, and IBM had agreed to pay for it.

Supercomputer Systems, Inc. opened in January 1988 with one product in mind. The SS-1 was meant to land roughly an order of magnitude past anything then shipping. IBM was the largest backer at more than $100 million, about half the total raised, with Boeing, Du Pont, Ford and the National Security Agency behind it in various combinations of money and interest.

The company grew past three hundred people. It never announced a delivery date it met, and the SS-1 slipped from the turn of the decade into 1992 and then out of 1992 as well. IBM had committed to fund it through the end of that year and then went further, covering weekly payroll into the new year while it made up its mind.

In January 1993 IBM declined to renew. Tech Monitor carried it on the 24th of that month, and the company retained a consulting firm to look worldwide for a replacement investor, which is the step before the end rather than a step away from it. SSI went under in the first months of 1993 and more than three hundred people in a Wisconsin city of under sixty thousand lost their jobs. There is no honest distance to write about a machine that shipped to nobody. Everyone on earth stood the same distance from the SS-1, including the people who built it.

~ the chronicle ~

~ the hundred million ~

The interesting thing about IBM's money is what it was buying. IBM was not short of computer designers in 1988. What it did not have was a vector supercomputer that could stand next to a Cray, and buying access to the man who had designed two of them was cheaper than starting an internal program from nothing. Then IBM's own parallel work started producing results, most visibly in what became the RS/6000 SP, and the argument for keeping an outside team in Wisconsin got weaker every quarter. The five-year commitment ran out in January 1993 and IBM chose not to write another one.

~ where chen went ~

Chen tried to salvage it immediately. SuperComputer International was formed later in 1993 to carry on the work and renamed Chen Systems in 1995. Sequent Computer Systems bought it in 1996, and IBM bought Sequent in 1999, so the engineering eventually went home to the company that had stopped paying for it six years earlier.

By 1999 Chen was in Shenzhen running Galactic Computing and building blade systems, which is close to the opposite bet: instead of one enormous custom machine, a great many ordinary ones in a rack. That bet was the right one and it was not his alone. The machines that won the 1990s were assembled from parts anybody could order, and the custom vector supercomputer turned into a thing you read about rather than a thing you specified.

The SS-1 has no museum piece because there was nothing finished to preserve. Denelcor at least shipped HEP machines that people wrote papers about, and Thinking Machines put a wall of blinking red lights into Jurassic Park. SSI produced five years of engineering and a bankruptcy. Its surviving record is other people's recollections and a few trade paragraphs from January 1993.

@}-->--

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

« previous  ·  random  ·  next »