Cyrix

» see cyrix.com as it lived, on the wayback machine

~ the obituary ~

The number printed on the box was not the speed of the chip. A Cyrix 6x86 sold as a P166+ ran at 133 MHz, and Cyrix said this was the honest way to label it, because on the work people actually did the part finished ahead of a Pentium clocked higher. The industry called it a Performance Rating. Intel and AMD called it a way to sell a slower chip. Both readings were true at the same time, which is most of the Cyrix story compressed into one line of a specification sheet.

The company started in 1988 in Richardson, Texas, founded by Jerry Rogers and Tom Brightman out of Texas Instruments. It sold math coprocessors first, the FasMath 83D87 in November 1989, then moved to whole processors that dropped into sockets people already had. The Cx486SLC arrived in March 1992 and the Cx486DLC that June. Cyrix owned no fabrication plants and never did, which meant every chip it designed had to be made by somebody else on somebody else's schedule. Intel spent years suing over the 486 work and lost, and ended up paying Cyrix 12 million dollars to settle the antitrust side of it.

The 6x86 shipped in 1995 and it was a good chip. Reviewers found it running 30 to 40 percent ahead of a Pentium at the same clock on ordinary Windows software. Then id Software released Quake in 1996, with its critical path written in assembly around the specific floating point pipeline of the Pentium, and a 6x86 could barely hold 15 frames a second unless you dropped the resolution to 320 by 200. Every hardware site on the web benchmarked with Quake, because Quake was what people wanted to know about.

A home PC in Turkey in 1998 was a parts list read out across a shop counter. You did not buy a machine, you bought a list, and every line on it was a place to save money. Cyrix was one of those lines. What you were deciding, without anybody putting it that way, was whether the thing you would spend the most hours doing on that computer looked more like a spreadsheet or more like a first-person shooter.

~ the case file ~

~ what the rating actually measured ~

The Performance Rating was not a lie so much as a claim about which benchmark counted. Cyrix built a wide integer core and spent its transistor budget there, so spreadsheets and the Windows interface itself ran faster than the clock number suggested. The floating point unit got what was left over and was not pipelined the way Intel's was. If your software did integer work, the rating was conservative. If it did math, it was fiction.

That was a defensible position in 1995 and an untenable one by 1997, because the software people used to decide what to buy had shifted underneath it. Intel understood the point well enough to launch the Celeron in 1998 at the prices Cyrix had owned, while keeping the family name attached to the number consumers already trusted. Competing on a rating meant explaining yourself at every point of sale, which is an expensive thing to have to do forever.

~ the quake problem ~

Quake was the first mass-market game whose performance came down almost entirely to floating point throughput, and its inner loop had been hand-tuned for the Pentium. On a 6x86 the frame rate collapsed. Cyrix could point out, correctly, that the chip was ahead in office software and ahead in most other games of the period, and none of that mattered, because the review a buyer read was the one with the Quake chart in it.

There is a version of this where the company survives by aiming somewhere else, and it half tried. The MediaGX in 1996 folded the graphics and the sound controller onto the processor die and sold into cheap prebuilt machines, which is a real market and one Cyrix briefly had to itself. It was also a market with no margin in it and no route back up to the parts that carry a brand.

~ what national semiconductor did with it ~

National Semiconductor completed its merger with Cyrix on 11 November 1997 and got what it wanted, which was the MediaGX and a way into low-cost integrated systems. The high end went unfunded. Through 1998 and into 1999 the MII crept from a PR-300 rating to PR-333 while AMD and Intel went past 450 MHz for real, and the last chip to carry the Cyrix name, the MII-433GP, ran at 300 MHz. The design engineers left one at a time and nobody replaced them.

National signed the definitive agreement handing the Cyrix PC processor business to VIA Technologies on 3 August 1999, after weeks of reporting that the deal was already done. By then there was no design team to transfer and no market left for the MII. VIA put the Cyrix name on a processor designed by Centaur Technology, shipped it as the VIA Cyrix III in February 2000 purely for the recognition, and dropped the name in 2001 when the same silicon became the VIA C3. Centaur's Glenn Henry gave the shortest available version of what had happened: Cyrix had a good product, then got bought by a big smokestack company and got bloated.

@}-->--

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

« previous  ·  random  ·  next »