Cydrome

1984 - 1988
dead us
summer 1988, and prime computer was set to buy the whole company. prime's board changed its mind at the last minute, which finished cydrome four years after it started and one year after its only product shipped. nine cydra-5 systems were ever built and three of those were prototypes. the architecture left with its chief designer for hewlett-packard, and turned up a decade later as the thinking behind itanium.

~ the last words ~

Cydrome began in San Jose in 1984 under the name Axiom Systems and then found out there was an older Axiom down in San Diego, so it went shopping for a name. What it had to sell was a word. It had been calling its architecture SPARC, and it sold the rights to that name, though not to the architecture, to Sun Microsystems. The proceeds paid NameLab to invent a replacement, and NameLab welded cyber to drome, as in a racecourse, to get Cydrome.

Everything here happened before I was online and most of it before I could read. The machine shipped in 1987, six production units of it, at a price only a corporate purchasing department would consider. I met these ideas backwards, through compilers that reorder your instructions without being asked, which is the only route into them for anyone who started out on a PC in the late nineties.

The founding group was Bob Rau as chief architect, David Yen running hardware engineering, his brother Wei Yen running software, along with Ross Towle and Arun Kumar. They moved out of San Jose to a business park in Milpitas on Presidents' Day in 1985. The machine was finished in 1987 and made its first public appearance at the inaugural Supercomputer Conference in Santa Clara that year.

~ the record ~

Born1984 in San Jose, under the name Axiom Systems
Peakthe Cydra-5 in 1987, at half a supercomputer's speed for a twentieth of the price
Killed1988, when Prime Computer's board killed the acquisition
Lifespanabout 4 years, 9 machines
Made byBob Rau as chief architect, with the Yen brothers running hardware and software
Killed bya board meeting at a customer

~ nine machines ~

The Cydra-5 was sold as a departmental supercomputer, which meant a price from half a million dollars up to a million for roughly half the performance of the ten to twenty million dollar machines of the day. Two kinds of processor shared the memory and the peripherals. A numeric processor did the mathematics using Cydrome's directed-dataflow design, a variant of very long instruction word, while an ensemble of Motorola 68020s ran Unix System V and handled everything that was not arithmetic. The operating system covered both, so the user saw one computer.

The numeric side used a 256-bit instruction word divided into seven fields, and the compiler's job was to find operations that could run at the same time and pack them into one word. It had rotating registers to make software pipelining of loops work, and it had an instruction cache but deliberately no data cache, on the view that a data cache is wasted on sparse arrays. Memory was interleaved 64 ways across four ports with the addresses hashed so that no part of it ran hot. All of it was ECL at 25 MHz. The Fortran compiler was the real product here, because the hardware could not schedule anything by itself.

Nine systems were built in total, six of them production machines and three prototypes. Prime Computer, which had invested in Cydrome late, resold the Cydra-5 under its own name. The Prime units were finished in black panels and Cydrome's own were white, which is the entire difference between them.

~ the summer of 1988 ~

In the summer of 1988 Prime was ready to buy the company outright, and at the last minute its board decided not to. Nothing replaced the deal. Cydrome closed that year after roughly four years of operation, having shipped single-digit quantities of a machine that worked.

Prime itself was in no position to be generous that summer. It was fending off a hostile bid from Bennett LeBow and had just bought a CAD company with borrowed money, and soon after that it went private in a leveraged buyout worth 1.3 billion dollars. The board that walked away from Cydrome was reading its own balance sheet, not a compiler manual.

~ what the compiler kept ~

Bob Rau joined Hewlett-Packard in 1989 and ended up an HP Fellow running the compiler and architecture research group at HP Labs. The work there produced Explicitly Parallel Instruction Computing, the architectural style underneath IA-64. He took the 2002 Eckert-Mauchly Award and died in December of that year. Wei Yen went on to found ArtX out of ex-Silicon Graphics engineers and later sat on ATI's board. The MIPS R8000 was heavily influenced by the Cydra work as well, so these ideas reached customers inside somebody else's processor years before Itanium did.

One Cydra-5 sits in storage at the Computer History Museum, catalogued under an accession number. cydrome.com now redirects to a domain broker's listing page.

~ 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.

paid once through stripe. payments are handled by OÜ Lakeview, the small company that keeps rip.so running. the money pays for 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: 3DR Solo, Celerity Computing, ZX Spectrum, Dell Adamo

also died in the 1980s: Celerity Computing, Casio PV-1000, Systime Computers

« previous  ·  random  ·  next »