iYAN inside! Online Index Hardware
Bye Bye Cyrix
A Market Innovator now bites the dust.
This news came as no surprise. National Semiconductor is exiting the PC socket processor business, focusing on integrated systems and possibly post-PC era computing devices. With the AMD vs. Intel price war at full blown, there is not even a price shelter where Cyrix MII CPUs can take refuge from the free fall of AMD K6-2 and Intel Celeron prices. In what seems to be one of the darkest quarters in the battle against Intel monopoly, AMD took its highest quarterly losses so far, and now Cyrix is throwing in the towel.
In one way, it was perhaps a greater innovator than AMD. It can be said that AMD never really an indigenous design of its own other than the 29000 RISC processor and the K5, which was a market failure. AMD relied on its licenses with Intel to make clone X86 processors from the 8088 all the way to the 486. The K6 was a second generation NexGen 586 design, since AMD acquired NexGen, and the K7 is the third generation. Cyrix began not with licenses or with acquisitions, but with "technology the hard way"---reverse engineering. Cyrix today may have a reputation for its lame floating point performance with its CPUs, but it started its business with a reverse engineered numeric coprocessor for the 286 and 386 that was supposedly better than the Intel 287 or 387. That was back in the days when the floating point unit was separate from the processor, and occupied another socket. While AMD punched out clone 386s and 486s using Intel microcode (and fought countless lawsuits with Intel), Cyrix continued to reverse engineer 386s and 486s. Technically, they were successful, but they were not faster or in anyway performed better, although the Cyrix 486 FPU actually outperformed the Intel 486 FPU by a small margin. It seems the the public trusted the Intel name more, or at least, would prefer an alternative CPU that got the genuine Intel microcode on it, like AMD or IBM Blue Lightning CPUs. Cyrix did carve out a niche though for 386 upgrades, using processors with 486 level performance as a drop in replacement on 386 sockets. When the Pentium era came, Cyrix extended the strategy by introducing early Pentium performance level CPUs as drop in upgrades to 486 sockets, sort of like 486s on steroids.
By then two patterns has emerged. The first pattern is the fabless company. Cyrix itself has no fabricating factories, relying on partners like SGS and Texas Instruments to manufacture the chips for them. The second is to establish a market niche by enhancing the life of the current PC platform by introducing faster upgrades for that platform. The green heat sink of Cyrix 486s and 586s (M1) became a trademark of these upgrades.
In my recollection of Cyrix, it was Epson that marked one of their earlier major OEM wins. Epson don't make PCs like they used to, but at that time, they were pretty serious. Opening up an Epson mini tower, one will be greeted with the discover of a Cyrix 486 or 586 with its attractive, shiny green heat sink.
The Pentium era became the litmus test for the alternative X86 world. No longer can a manufacturer rely alone on licenses of 386 or 486 microcode to keep up. You're on your own. Not since the RISC based 29000, did AMD had to design a new processor from scratch. Combining a few RISC techniques, AMD created its first reverse engineered design, the K5. The results was mixed. The K5 succeeded in its basic goal, compatibility with Pentium Socket 5, as well as with X86 software. Clock to clock it was even faster than the Pentium. On the other hand, AMD can't deliver the product in time and in quantity, a reflection of its production snafus to come, and the K5 cannot keep up with clock speed advances of the Pentium. Worst was that the K5 could not compete with a new rising star, the 686 from Cyrix.
At the end of the rope, AMD halted its K5 program, and development of its own internal K6 design. Instead, it bought a company called NexGen, which introduced a novel idea to X86 architecture. That idea was to use a RISC core which will execute internal RISC opcodes, with a front end decoder that translates X86 opcodes to the internal RISC opcode format. NexGen, like Cyrix, was originally a fabless company. It's novel idea was no vapor; it did deliver its NexGen 586 chip, as well as its specialized cache controller, and there were motherboards made for the NexGen. AMD absorbed NexGen, its talents went on to begin the "next generation" NexGen processor, which ultimately became the marketed AMD K6.
It was here on the Pentium era, that the Cyrix talent of reengineering paid off. The breakthrough product was something originally code named M2, and it became popularly known as the 686. At this point, Intel scuttled the X86 labeling format, since apparently names based on numbers cannot be copyrighted. The predecessor, M1, wasn't taken seriously as a Pentium competitor, but like Cyrix products before, viewed as a "chip" upgrade to older 486 platforms. Using techniques learned from RISC designs, the M2 was the first alternative X86 chip to outperform a contemporary Intel processor, this time, the Pentium. It's performance puts it in between the Pentium and the much more expensive Pentium Pro, and for a price cheaper than both.
With the M2, Cyrix has arrived. It's got the backing of IBM, who would manufacture the 686s and use them for its own PCs. It's price/performance is unmatched. For the year 1996 and 1997, it would actually outsell AMD on alternative X86 chips. It would change the X86 market from an almost has been to a serious competitor. With its low prices, Cyrix would also father the sub $1,000 PC market. It's a revolution. Cyrix was at the top of the world.
My own personal involvement with Cyrix began with the acquisition of a 686 PR166. My first ever piece was an impressive physical sample. I have never seen a package with its hefty weight and gold finish with the exception of the much more expensive Pentium Pro. This was one of those early types that ran hot, which gave Cyrix the reputation of aggressive temperatures that it never fully outlived. It was matched to a motherboard made by Microstar with an Intel 430HX chipset. All my processors have been Intel or AMD before that, and my AMD chips have included licensed designs like 8088, 286, 386 (especially the beloved 40MHz part) and 486. At least three of my office machines had K5s, two of them, retired only by mandate recently. They and the other one still working could have gone on forever.
I had also migrated my OS/2 platform from a 486DX2-66 to a Pentium 75, then a K5 as an experiment, and finally to the 686. The transition felt like OS/2 shed its shackles off its feet. It flew like OS/2 and Cyrix was made for each other. I moved the 686 to a Windows 95 machine with a FIC PT2006 motherboard, which was one of the most impressive motherboards based on the Intel 430VX chipset. The Microstar was becoming flaky for some reason, probably hit by a power surge, and was retired after providing me a period of good service. The system was a dual partition one, with OS/2 on one side and Windows 95 on the other. I do use the machine to play games. Somehow, the weak FPU of Cyrix wasn't a factor to the games I prefered playing then, including the Mechwarrior 2 series, Privateer 2, EarthSiege 2, Mission Force: Cyberstorm, Master of Antares 2, and under OS/2, Galactic Civilizations 2. But the heavier overhead of an advanced operating system like OS/2 requires strong integer and general application performance, and the Cyrix fits the bill. The fact that some of the chips are labled "IBM" added some poetic justice to be used with the IBM operating system.
At the same time, I have also begun experimenting with AMD K6 CPUs, first on an Asus T2P4 motherboard, and then later a TX97-E motherboard. I wasn't impressed with the Pentium 200 and Pentium 200MMX, as they cost too much for their relative performance. The K6s performed better than the Pentiums, but they are still out priced by the Cyrix chips, which performed equal to better. Cyrix systems were quickly spreading in our offices.
FIC was the motherboard brand I choose with my Cyrix. But this was not before I had two Cyrix systems on non FIC boards, one on a CTX branded PC which is a 686L PR200, and another on a home made PC using a Lucky Star motherboard with an Intel 430TX chipset, this one a 686MX-PR166. To bless my own office OS/2 system, I took the most highly regarded socket 7 motherboard then, the VIA VP2 based FIC PA2007. This board took a sweep in benchmarks and reviews around the late 1997, especially on Tom's Hardware site. On top of that I placed a 686L PR200. Later changed to a 686MX, this system has become my trusted work machine. Many of my articles on this magazine was born on this machine, and sadly enough, it is on this Cyrix machine that I will be saying this eulogy to Cyrix.
Another office system was Cyrix-converted, with a 686MX-PR200 on an FIC PT2007 motherboard, which had an Intel 430TX. This conversion, however, proved to be one of the most baffling. OS/2 could not install successfully on this system, and the BIOS kept identifying the chip as a PR166, even though the jumper settings are correct for a PR200 rating. I was a bit of dim bulb with such matters then, and it's much clearer now in hindsight that the PT2007 needed a BIOS update that would properly recognize the processor as a PR200. As it was, it was on a genuine 166MHz clock cycle, which deserves the PR200 rating, but based on outdated specs, called it a PR166. PR or Performance Rating isn't true MHz rating, but it was quite confusing then. As a matter of fact, my Asus T2P4 motherboard also called the K6-200, "K6 PR200"
Nonetheless, the performance was eye opening. At that point in time, this was my fastest system to date. I had a 686MX PR166 which I bought rather cheaply as well, and I had decided to overclock it as installed on my FIC PA2007 motherboard to 75x2 (150MHz) for a PR200 rating. Actually, I merely dropped the 686MX without changing the jumpers, since it was already set for the previous CPU, the 686L PR200. Booting up however, my particular PA2007 showed the same symptom of an outdated BIOS that identifies my 686MX as a PR166 although the proper rating should be PR200. Somehow the PT2007 felt faster even than the PA2007, although the Intel TX chipset is hardly as optimized for the Cyrix compared to the VIA chipset. Although the PT2007 is set at 66x2.5, changing the clock settings between 75MHz x 2 and 66MHz x 2.5 (same PR rating as I shall discuss below), didn't seem to matter. In fact, the Cyrix should be a bit faster with the faster bus speed as opposed to the core speed.
VIA isn't known for its Intel chipset products, and the PT2007 is a virtually ignored entrant. This is in contrast to FIC's reputation with VIA chipsets and the PA2007 is a great seller. It turns out the only other confirmation that the FIC PT2007 was actually a little bit faster with the Cyrix than the PA2007 proved to be from a rising website called Anandtech. From then on, because of this little detail, I became a great fan of this website, which still hosts in its archives, the greatest collection of motherboard reviews anywhere in the web.
The PR rating has proven to be a great source of confusion. What more, BIOSes often try to guess ahead the anticipated bus x clock jumper settings for unreleased PR settings. When Cyrix finally announces the settings for the higher PR ratings, the BIOSes tend to get them all wrong, requiring a round of BIOS updates to set CPU ID right. On the running part, there is really is no difference at all. Your software will run no faster or slower. It's just neat for your morale to see the proper PR rating when boot up.
