Supertek Computers

1985 - 1990
dead us
mike fung left hewlett-packard in 1985 to build a cray you could plug into a wall socket. the supertek s-1 ran cray x-mp code and fit in a room with no chilled-water loop, at about a quarter of a million dollars. ten of them sold. cray research bought the company in 1990 and shipped the design under its own badge.

~ the obit ~

A Cray X-MP in 1985 was ECL logic and a liquid cooling system that had to be engineered into the building around it. It cost several million dollars, and the people who could authorize that purchase mostly worked for governments. Supertek Computers was founded that year in Santa Clara on the idea that the same instruction set could be built in CMOS, cooled with fans, and sold for a fraction of the price.

Mike Fung had been a project manager at Hewlett-Packard before he started it. The company raised $21.4 million in venture capital and spent about $5 million of that developing the first machine, which is an unusually good ratio and turned out not to matter. In May 1989 Cray Research and Supertek announced a royalty-free cross-license on each other's patents, which is the sort of arrangement you make with a small competitor you have decided not to sue.

The S-1 shipped in 1989 at around 36 megaflops and about $250,000. It ran Cray's own operating system and later Unix, so software a customer already owned would run on it without a port. Ten were sold. In 1990 Cray Research bought the company outright, and Supertek stopped being a name five years after it became one.

~ the rap sheet ~

Born1985, santa clara california
Peakthe s-1 in 1989, ten units sold
Killed1990, acquired by cray research
Lifespan5 years
Made bymike fung, formerly a project manager at hewlett-packard
Killed byten machines sold against $21.4 million of venture capital

~ selling a cheaper cray ~

The pitch was sound on paper. Far more organizations wanted to run Cray software than could afford to buy a Cray, and the gap between those two numbers was the entire minisupercomputer industry. Convex and Alliant both sold into it, and a smaller group went further by making their machines binary-compatible with Cray rather than merely fast.

Compatibility is a harder promise than speed. A machine that claims to be fast needs benchmarks. A machine that claims to be an X-MP has to be wrong about nothing, because the customer's code was written against the real one and nobody is going to debug your emulation for you. Supertek made that promise in CMOS in 1989 and appears to have kept it. The processor was never the problem.

~ the ten customers ~

Ten units against $21.4 million of venture money is the whole reason the company was sold rather than grown. It is not a failure of the machine so much as a failure of the count. The market for a cheaper X-MP existed, but it was a market of dozens rather than hundreds. A hardware company needs hundreds to pay for the next design. Denelcor sold six HEPs and died. Multiflow sold considerably more and died anyway, while Supertek got itself bought, which on this part of the site counts as the good outcome.

~ what cray did with the design ~

Cray Research sold the S-1 for a short while as the Cray XMS, from 1990 into 1991. It did not go well, and the people who worked on it are fairly clear that the trouble was the input/output subsystem rather than the processor. The IOS was built on VME, and VME could not move data fast enough to keep the machine busy on real work.

The S-2, a Y-MP-compatible follow-on still in development when Cray bought the company, had a much longer life. It arrived in 1992 as the Cray Y-MP EL, the entry-level model, and that line ran on into the J90 series. A CMOS reimplementation of a Cray architecture turned out to be exactly the right idea. It needed a sales force and a support organization behind it, which is what Supertek was buying when it agreed to be acquired, whether or not that was the plan going in.

Supercomputing did get cheap in the end, but by a different route. Ordinary microprocessors got fast enough that you could put a lot of them in a rack, and that is the trend which closed every company on this part of the site. That version reached Turkey. University clusters built out of commodity boxes were around by the time I was finishing my degree, and they descend from Beowulf rather than from anything built in Santa Clara in 1989.

~ leave a rose ~

an anonymous tribute. one rose per visitor per day.

@}-->-- ... roses left here

~ leave a wreath ~

a rose fades by morning. a wreath stays on this grave for good, with your name on it.

from $1, paid once through stripe. it keeps the graveyard running: the hosting and the nightly digging. adopt it and the headstone reads "maintained in memory by" you. the groundskeeper plants new wreaths at the next tending, usually within a day.

» one email a week: new graves and the anniversaries of the dead. subscribe

~ leave a note ~

visitors before you have left these graveside notes. anonymous welcome.

other gadgets: Handy (Tink Labs), Hello Sense, Macintosh TV, Google Glass

also died in the 1990s: Hewson Consultants, Microsoft NetShow, Macintosh TV

« previous  ·  random  ·  next »