The Quick Cyrix Guide
Founded in 1988 by a number of former TI-ers, Cyrix was a fabless processor designer. Their initial products were x87-compatible math coprocessors, but they soon began designing complete x86-compatible processors. Their life as a company was tumultuous and relatively brief; constantly hounded by legal issues with Intel and problems with its manufacturing partners, as well as issues keeping up with the performance of competing products, Cyrix ultimately merged with National Semiconductor in 1997. National Semi held on to the remains of Cyrix for a few years, then sold most of Cyrix's intellectual property to VIA in 1999, eventually selling the rest to AMD in 2003.
The History
Cyrix got its start in 1988, selling math coprocessors (floating-point units) intended to compete with Intel's 80287 and 80387. The Cyrix "FasMath" parts outperformed their Intel counterparts by a decent margin, and helped the company build an initial foothold.
As Intel moved into the 486 era, Cyrix presumably looked at the 486DX's integrated floating-point unit and saw the handwriting on the wall. The coprocessor niche would disappear a few years later, as every Intel processor from the Pentium onward included a floating-point unit onboard as standard equipment. Thus it was that Cyrix began to move into the x86-compatible processor market.
Cyrix was unfortunately not in the same position as AMD. AMD had previously been a contract manufacturer of Intel's products, which gave them access to the complete designs and microcode for the 286. For subsequent products, AMD still had to build clean-room designs that acted like Intel's products but did not use the same microcode, but having so much information available made that task much easier. Cyrix was put in the less comfortable position of reverse-engineering Intel's products with no special knowledge.
The result was that while AMD's products achieved excellent compatibility with the Intel instruction set, Cyrix could not reach the same level of compatibility. Early Cyrix products especially were known for significant compatibility issues, and consequently even today the specter of non-Intel products having potential compatibility problems haunts every non-Intel processor.
Despite all of this, Cyrix faced a major lawsuit from Intel disputing Cyrix's right to make x86-compatible processors. Intel claimed that Cyrix was violating Intel patents. The lawsuit was eventually settled, with Cyrix retaining the ability to make x86-compatible processors, but only at foundries that already had an Intel license.
Cyrix was a fabless company and, probably because so many of its employees had come from Texas Instruments, its original fabrication partner had been Texas Instruments. But there were major problems with this relationship, not least that Cyrix had trouble getting priority at TI's best fabs. This was a major contributor to Cyrix's long-term performance problem: although their small team was able to design decent products, the yields of high-frequency parts were never good. Thus they could do well on a clock-per-clock basis, with a high score in instructions per clock (IPC), but due to low clock frequencies their performance still lagged behind Intel. Cyrix would eventually turn to SGS Thompson (whose fabs were later spun off as STMicroelectronics) and then IBM for foundry services.
The deal with IBM looked like it might save the company, but it probably was the last nail in Cyrix's coffin. Under the terms of the deal, IBM would open its very advanced fabs to Cyrix, but would also be allowed to sell Cyrix designs under its own brand. Many in the industry expected that IBM would use the resulting products in many of its own computers, possibly strengthening Cyrix's reputation. Instead, IBM used them in only a handful of its computers, and primarily sold them into the retail market. This cut directly into Cyrix's own sales, particularly since IBM sometimes priced its products below the equivalent Cyrix part.
Under the weight of its legal, manufacturing, and performance problems, Cyrix agreed to merge with National Semiconductor in a deal that went through in 1997. This was generally seen as a good move; the merger gave Cyrix access to National Semi's fabs, capital, and name recognition. But National Semi had financial troubles soon after, and did not invest much money in the microprocessor market. In 1999, they unloaded most of Cyrix's remains to VIA. They held on to the MediaGX until 2003, when that was sold to AMD and reborn as the AMD Geode.
The Technology
Cyrix had some significant accomplishments, all told, even though those accomplishments were never enough to give them a strong position in the market.
Probably their greatest achievement was reverse-engineering compatibility with the x86 instruction set without any of the inside knowledge that AMD had. This was hardly a trivial task, particularly given that x86 is one of the most complex instructions sets of any mainstream microprocessor family. But the imperfections in their reverse-engineered designs, particularly the earliest ones, branded them with a reputation as the maker of cheap imitations, and they were never able to shake that label.
Cyrix pioneered power management and register-renaming techniques. They had good reason to be interested in power management; for most of their history, Cyrix products lagged significantly behind Intel and AMD on process node changes. Because their products were often made on older, larger process nodes, those products produced more heat than their competitors. This was likely compounded by the fact that later Cyrix products often relied on increased bus speeds to boost performance (meaning that while the clock frequency was lower, more parts had to run at the full clock frequency).
The Products
Cyrix's first products were the 486SLC and 486DLC, which were intended as upgrades for owners of 386-based systems. They were pin-compatible with the 386SX and 386DX, respectively. Despite the name, they really sat somewhere between a 386 and 486; they were hybrid 16/32-bit parts (like the 386SX), but had on-board L1 cache (unlike the 386, although they had less than the 486).
Cyrix then went on to produce its own take on 486DX/2 and 486DX/4 parts, with less than stellar results. Slightly ironically, floating-point performance of Cyrix parts was fairly weak.
Cyrix's next product on the market was the 5x86 (sometimes written Cx5x86 to distinguish it from the remarked SGS-Thompson and IBM versions), which was intended as an upgrade for owners of 486 systems. It was actually a cut-down version of the 6x86 (the "M1" core), which would come out a few months later. Compared to a 486, it was quite a performer; compared to the Pentium, not so much.
The 6x86 was released to the market a few months later, in December 1995. The 6x86, like the K5 and Pentium Pro, had a RISC-like core, and x86 instructions were translated to an internal language before execution. Notably, the 6x86 was introduced very shortly after the Pentium Pro, and before the K5, despite the tiny design team that Cyrix had to work with. Performance was good and the price was low.
In 1997, Cyrix released the first of its MediaGX line, which would both save and doom the company. The MediaGX was a highly integrated product, approximating what would now be called a system-on-chip (SoC). It housed a core based on the 5x86, video, audio, and PCI bus controller. It was intended for embedded applications where performance was not a priority, and cost was. Cyrix was eventually swooped up by National Semiconductor largely because National Semi wanted the MediaGX; yet, a shift in focus toward the MediaGX and away from the mainstream market was part of what had caused Cyrix to falter in the first place.
Also in 1997, Cyrix released the 6x86MX (M2 core), which added support for Intel's MMX instructions and a larger L1 cache. Performance was still strong, as this was essentially just a beefier version of the 6x86.
The 6x86 would, in later iterations, morph into the "MII". This was the same product, but a different name. The name change was presumably meant to help the product stand next to the Pentium II and K6-2 that were coming on to the market. The MII was first announced in 1998, and would be the last product Cyrix introduced before merging into National Semi, after which they released only new iterations of the same products.
Editor's note: This article is part of the expansion plan for the Big Processor Guide. It will make a lot more sense in the near future.