Cyrix: Gone But Not Forgotten

Most of you are no doubt familiar with Intel and AMD, Qualcomm, Texas Instruments, and possibly even VIA – but there's another precursor chip maker that you should be familiar with.

For the better part of a decade, Cyrix brought the world of personal computing to millions in the form of attainable budget PCs, only to be killed by its best product and its inability to run a popular game, followed by a bad merger with a larger partner.

The early 1990s were a strange time for the desktop computing industry.

It looked like Intel was winning despite fierce competition in the microprocessor space; Apple switched to IBM's PowerPC architecture, while Motorola's 68K chips were slowly dragging Commodore's Amiga PC to the grave. Arm was only a tiny flame sparked by Apple and a few others, and was almost entirely focused on developing a processor for the infamous Newton.

This was around the same time AMD was liberating its processors from the negative aura of being second-sourced from Intel. After cloning a few more generations of Intel CPUs, AMD came up with its own architecture, which by the end of the nineties were well regarded in terms of price and performance.

That success can be attributed at least in part to Cyrix, a company that had a window of opportunity to capture the home PC market and leave both Intel and AMD in the dust, but ultimately failed to execute and quickly disappeared into the tech graveyard.

Modest Beginnings

Cyrix was founded in 1988 by Jerry Rogers and Tom Brightman, starting out as a manufacturer of high-speed x87 math co-processors for 286 and 386 CPUs. These were some of the greatest minds to leave Texas Instruments and they had high ambitions to take on Intel and beat them at their own game.

Rogers embarked on an aggressive pursuit to find the best engineers in the US and proceeded to become an infamously hard-driving leader for a team of 30 people that were tasked with the impossible.

The company's first math coprocessors outperformed Intel equivalents by ~50% while also being less expensive. This made it possible to pair an AMD 386 CPU and a Cyrix FastMath co-processor and get 486-like performance at a lower price, which caught the industry's attention and encouraged Rogers to take the next step and pursue the CPU market.

In 1992, Cyrix unveiled its first CPUs, the 486SLC and 486DLC, which were intended to compete with Intel's 486SX and 486DX. They were also pin-compatible with the 386SX and 386DX, meaning they could be used as drop-in upgrades on ageing 386 motherboards, and manufacturers were also using them to sell budget laptops.

Both variants offered slightly worse performance that an Intel 486 CPU but significantly better performance than a 386 CPU. The Cyrix 486 DLC wasn't able to compete with Intel's 486SX offering clock-by-clock, but it was a fully 32-bit chip and sported 1KB of L1 cache, while costing significantly less.

At the time, enthusiasts loved the fact that they could use a 486DLC which ran at 33 Mhz to achieve comparable performance to that of an Intel 486SX running at 25 MHz. That said, it wasn't without problems, as it could lead to stability issues for some older motherboards that didn't have extra cache control lines or a CPU register control to enable or disable the on-board cache.

Cyrix also developed a "direct replacement" variant called Cx486DRu2, and later on in 1994 released a "clock doubled" version called Cx486DRx2, which had the cache coherency circuitry integrated into the CPU itself.

By then, however, Intel had released its first Pentium CPU, which drove 486DX2 prices down to the point where the Cyrix alternative had lost its appeal as it was cheaper to upgrade to a 486 motherboard than it was to buy a Cyrix upgrade processor for an old 386 motherboard. When the "clock tripled" 486DX4 arrived in 1995, it was too little, too late.

Large PC manufacturers such as Acer and Compaq weren't convinced by Cyrix's 486 CPUs and instead opted for AMD's 486 processors. This still didn't stop Intel from spending years in court alleging that Cx486 violated its patents, without ever winning a case.

Cyrix and Intel eventually settled outside of court and the latter agreed that Cyrix had the right to manufacture its own x86 designs in foundries that happened to hold an Intel cross-license, such as Texas Instruments, IBM, and SGS Thomson (later STMicroelectronics).

Never Repeat the Same Trick Twice... Unless You Are Cyrix

Intel launched the Pentium processor in 1993, based on a new P5 microarchitecture and finally coming up with a market-friendly name. But more importantly, it raised the bar in terms of performance that ushered in a new era of personal computing. The novel superscalar architecture allowed it to complete two instructions per clock, a 64-bit external data bus made it possible to read and write more data on each memory access, the faster floating point unit was capable of up to 15 times the throughput of the 486 FPU, and several other niceties.

Cyrix took on the challenge to yet again create a middle ground for Socket 3 motherboards that were not able to handle the new Pentium CPU, before that model was even ready to ship. That middle ground was the Cyrix 5x86, which at 75 MHz offered many of the features of fifth-generation processors like the Pentium and AMD's K5.

Cyrix 5x86 CPU with heatsink. Image: NostalgiaNerd

The company even made 100 MHz and 133 MHz versions, but they didn't really have all the advertised performance-enhancing features since they would cause instability if enabled, and overclocking potential was limited. All of these were short-lived and in six months Cyrix decided to stop selling them and moved on to a different processor design.

Peak Cyrix Through the Lens of Quake

In 1996, Cyrix unveiled the 6x86 (M1) processor, which was expected to be yet another drop-in replacement for older Intel CPUs on Socket 5 and Socket 7 motherboards with decent performance. But this wasn't just an upgrade path for budget systems, it was actually a little marvel in CPU design that was thought to do the impossible – it combined a RISC core with many of the design aspects of a CISC one. At the same time, it continued to use native x86 execution and ordinary microcode, while Intel's Pentium Pro and the AMD K5 relied on dynamic translation to micro-operations.

The Cyrix 6x86 was pin-compatible with the Intel P54C and had six variants with a confusing naming scheme that was supposed to indicate the expected performance level, but wasn't an actual indicator of clock speed.

For instance, the 6x86 PR166+ only ran at 133 MHz, and was marketed as being equivalent to or better than a Pentium running at 166 MHz, a strategy that AMD would replicate later on.

Be that as it may, the problem was that the 6x86 actually identified itself as a 486 CPU because it didn't support the full Intel P5 instruction set. This would quickly become an issue as most application development was slowly migrating towards P5 Pentium-specific optimizations to squeeze more performance using the new instructions.

Cyrix eventually improved compatibility with the Pentium and Pentium Pro through the 6x86MX and 6x86MII variants.

A huge selling point of the 6x86 was that its integer performance was significantly better than the Pentium's, which was a good advantage to have at a time when the vast majority of applications and games relied on integer operations. For a while, Cyrix even tried to charge a premium for that added performance, but eventually that strategy fell apart.

Cyrix 6x86MX CPU die shot

As it turned out, the FPU (floating point unit) of the 6x86 was only a slightly modified version of Cyrix's 80387 coprocessor, and as such, significantly slower than the new FPU design integrated by Intel's Pentium and Pentium Pro.

To be fair, it was still anywhere between two and four times faster than the Intel 80486 FPU, and the Cyrix 6x86 bested the Intel offerings on overall performance. But that whole equation broke down when software developers, particularly those making 3D games, saw the rising popularity of the Pentium and chose to optimize their code in assembly language around the advantages of the P5 FPU.

When id Software released Quake in 1996, PC gamers using 6x86 processors discovered they were getting sub-par frame rates that reached at most, an unplayable 15 frames per second, unless they wanted to drop the resolution down to 320 by 200, in which case you'd have needed a top of the line, Cyrix 6x86MX PR2/200 CPU to get a playable 29.7 frames per second. Meanwhile, gamers with Intel systems had no problem running the game at playable frame rates even at 640 by 480.

John Carmack had figured out that he could overlap integer and floating point operations on Pentium chips, as they used different parts of the P5 core for everything except instruction loading. That technique didn't work on the Cyrix core, which exposed the weakness of its FPU. Reviewers at the time found that in every other benchmark or performance test, the 6x86 CPU would leapfrog the Pentium by 30 to 40 percent.

Back in the mid 90s, no one knew the exact direction that computing would take, and Cyrix thought it was best to prioritize integer performance, so it produced a processor that lacked instruction pipelining, a feature that would become an essential part of a desktop CPU. Instruction pipelining is a technique that divides tasks into a set of smaller operations that are then executed by different parts of the processor simultaneously, in a more efficient fashion. The FPU of the Pentium processor was pipelined, leading to a very low latency for floating point calculations to handle the graphics of Quake.

The problem was easy to solve and software developers have released patches for their applications and games. But id Software had spent too much time designing Quake around the P5 microarchitecture and never provided such a fix. AMD's K5 and K6 CPUs fared a little better than Cyrix's, but they were still inferior than Intel's offerings when it came to Quake, which was a really popular game and a flagship among a new breed of 3D titles.

This had Cyrix CPUs becoming harshly judged on that performance gap, and the company all but lost credibility in the eyes of many enthusiasts. Because the company had been unable to score contracts with large PC OEMs, it was a particularly hard blow for Cyrix's fierce customer base that was made up of those same enthusiasts.

To make matters worse, Cyrix was a fabless chip maker that relied on third parties to manufacture its processors, and those companies used their most advanced lines for their own products. As a result, Cyrix processors were manufactured on a 600 nm process node while Intel's were 300 nm.

Efficiency suffered, and this is also why Cyrix CPUs had a reputation for getting extremely hot – so much so that enthusiasts were designing hotplates using them as a heat element. They were overly sensitive to low-quality power supplies, and their overclocking potential was also limited, but that didn't stop people (like this author, whose second PC had a Cyrix 6x86-P166+ CPU inside) from pushing them just a little bit and slowly leading them to their demise.

The Fall of the First True Rival to Intel's CPU Hegemony

By 1997, Cyrix had tried everything in their power to forge a partnership with companies like Compaq and HP, as integrating its CPUs into their systems would have generated a steady income stream. It also tried suing Intel for infringing its patents on power management and register renaming techniques, but the matter was settled quickly with a mutual cross-license agreement, so that the two firms could stay focused on producing better CPUs.

A famous National Semiconductor ad

The litigation took a toll on the already cash-strapped company. Faced with the prospect of bankruptcy, Cyrix agreed to be merged into National Semiconductor. This was seen as a blessing. The company would finally have access to proper manufacturing plants and a strong marketing team that was able to score large contracts. The IBM manufacturing agreements held on for a while, but Cyrix eventually moved all production to National Semiconductor.

Faced with the prospect of bankruptcy, Cyrix agreed to be merged into National Semiconductor.

Yet as it turns out this move would seal Cyrix's fate. National Semiconductor wasn't interested in making high performance PC parts, and instead wanted low-power SoCs (system on a chip).

Sure enough, Cyrix came up with the universally-hated 5x86 MediaGX, a chip that integrated functions like audio, video, and memory controller with a 5x86 core running at 120 or 133 MHz. It was a low performer, but it managed to convince Compaq to use it in their low-end Presario computers. This whet other OEM's appetite for 6x86 CPUs, with Packard Bell and eMachines as notable examples.

The shift in focus didn't stop Cyrix from trying to produce more high-performance CPUs, but it delivered promises and little else. National Semiconductor eventually sold Cyrix to Taiwan-based chipset maker VIA Technologies, but by then key people had already left and the MII CPU was an uninteresting part that found no buyers.

The last Cyrix design was the MII-433GP which ran at 300 MHz and, thanks to the unfortunate naming scheme, ended up in comparisons with processors that ran at 433 MHz, which were vastly superior. AMD and Intel were busy racing to 1 GHz and beyond, and it would take 20 more years for Arm to come along and challenge the two giants in the desktop and server markets – not to mention totally dominate mobile computing.

VIA put the final nail in the coffin as it used the Cyrix name to replace "Centaur" branding on processors that actually used an IDT-designed WinChip3 core. National Semiconductor kept selling the MediaGX for a few more years, before rebranding it into Geode and selling the design to AMD in 2003.

Three years later, AMD demonstrated the world's lowest-power x86-compatible CPU, which took only 0.9 watts of power and was based on the Geode core, a testament to the ingenuity of the Cyrix design team.

Why Cyrix's Legacy Matters

Whether or not you ever owned a Cyrix-powered PC, the company should be remembered for its legacy and lessons learned. Despite the relatively small influence on the industry during its decade of existence, Cyrix's failures proved that improving IPC (instructions-per-clock) was a more productive endeavor for chip makers compared to improving raw clock speeds.

To this day, Intel and AMD have tried to push nominal clock speeds higher with each generation, but after the 3 GHz milestone, most of the real improvements have come from rethinking core parts of their microarchitectures (and caching). A notable example is AMD's Zen progression, which has brought single-threaded performance improvements of 68% in less than four years.

Cyrix was able to survive and overcome a lot of legal (and by extension, financial) pressure from Intel, who sued almost everyone in the CPU space in the 1990s. It showed on two occasions that litigation is detrimental for a healthy marketplace while cross-licensing deals lead to a lot of cross-pollination between engineering efforts at different companies, which proved beneficial.

Cyrix also operated as a fabless company before that was cool. These days it's standard practice for most silicon giants, including the likes of AMD, Qualcomm, Broadcom, Nvidia, Apple, Marvell, Unigroup China, and HiSilicon, who depend on other companies to manufacture their chips.

Cyrix operated as a fabless company before that was cool.

The company's marketing strategy was never great before the National Semiconductor merger, and AMD would repeat the same mistakes with Athlon and Sempron processors in the 2000s.

These were labeled as to indicate that they were faster than an Intel processor, while operating at a lower clock speed, but that didn't always translate well in benchmarks or real-world performance tests. AMD dropped that scheme, but suffice to say, things remain a bit confusing to this day.

Today, it's unlikely you'll find a Cyrix processor outside of gold reclaiming operations and enthusiasts' vintage computer collections. There's some evidence online that Cyrix-based desktops were in use up until at least 2010, meaning they lingered for another decade after the company had essentially dissolved into VIA Technology's soup. It's unlikely that VIA's Zhaoxin arm still uses anything coming from the original Cyrix design, but only time will tell if they learned the lessons to honor Cyrix's legacy.