Whatever happened to them?
Back in the 90s, they were one of the leading developers of x86 compatible processors, but
you mention the brand now, and people might think you're talking about the best enhance
shaman on the face of the universe... if you can even stand on the face of the universe?
Wouldn't that be god's face?
Or maybe they might think you've mispronounced Citrix.
But a microprocessor developer is unlikely to spring to mind.
So let's begin at the beginning and follow this primarily 90s tale of semiconductor fabrication.
It's a sunny day in Richardson Texas, the year is 1987 and Tom Brightman and Jerry Rogers
are in a good mood.
Mainly because they've just set the seeds in motion for Cyrix Corporation.
The world of solid state components is accelerating rapidly and the duo believe they've identified
an area where they can thrive.
Neither is a stranger to this world, both having worked at Texas Instruments, before
following brimming career paths.
Jerry's stint at TI involved work as a VLSI design engineer and manager, filing 12 patents
for microprocessor and controller work.
He became Vice President and manager of the microprocessor division in 1981 before moving
to become CEO of Visual Information Technologies.
It was from here that he met Sevin Rosen, a venture capital firm that would assist in
the launch of Cyrix with $4 million of start up capital.
Tom also worked in the field of VLSI since 1975, designing numerous network, microprocessor
and DSP based system products and integrated circuits.
After Texas Instruments he was key in manufacture at both Commodore and Atari before reuniting
with Jerry in 1987 to start on their new venture.
The niche the duo had discovered was in the world of intel compatible math co-processors.
The market was ready and there was little competition other than Intel's own components.
But rather than x-raying Intel's chips and copying the layout, instead Cyrix based their
chips around completely new logic referring to the documented and seeking the undocumented
functions of Intel's products.
Using these methods and armed with only a 30 man design team, Cyrix were able to create
2 clones with a very high degree of compatibility, These were the Cyrix FasMath 83D87 and 83S87,
for the DX and SX 386 chips respectively.
Unlike the 387DX, the Cyrix chips do not support asynchronous operation of both the CPU and
co-processor but had advanced power saving features which automatically shut down unused
parts of the chip.
The increased reliance on hard wired logic over microcode, also allowed the Cyrix designs
to operate at a greater speed than Intel's implementations on floating point operations.
Cyrix being a fabless operation outsourced manufacture of these initial chips to their
friends at Texas Instruments, and by 1989, had launched their DX chip at both consumer
and business users, with the SX variant following in 1990.
Both chips slightly undercut Intel equivalents in price, making them a very tantilising prospect
for users looking to upgrade their machines with floating point acceleration, and initially,
this was the main selling point.
In the field, floating point benchmarks of the 83D87 showed performance almost twice
as fast as Intel's equivalent, although in other tests, this was much less pronounced.
But it was enough to secure Cyrix's place in the market and their expansion onto bigger
and better things.
This started with the EMC87 co-processor in June 1990, which offered even greater speeds
than their previous units, followed in 1991 by the FasMath CX-82S87 - such a catchy name.
Based around their 83D87, this was a 287 interchangeable coprocessor, although a much faster one, quickly
finding it's place as the fastest co-processor available for 286 based machines.
Cyrix weren't mainstream by this stage, but were quickly acquiring a name for doing things
faster and cheaper, so in 1992 when Cyrix announced upgrades for actual 386 processors,
it was no real surprise.
By now Intel's 486 processor was becoming well embedded in desktop computers, having
launched in 1989, however many users were still using 386 machines and wanted a cheaper
It's with this in mind that Cyrix released the 486SLC and 486DLC chips in May and June
The SLC was a pin compatible replacement for 386SX processors, which only had a 16 bit
data-path, with the DLC a replacement for DX equivalents, offering a full 32 bit data-path
Available in 25, 33 and 40MHz flavours, benchmarks were quite favourable, especially when cost
was considered, with the 40MHz DLC able to outperform a true 25MHz 486SX machine.
Essentially these chips contained a 486 instruction set, coupled with a 1KB Level 1 cache - 7KB
less than an Intel 486, and quickly acquired the moniker of, "a half 486".
But they weren't the only processors of this kind.
386 upgrade competition was also provided by Intel's own RapidCAD processor and Chips
& Technologies Super386, however both of these lacked even the 1KB of L1 cache and proved
less successful than the Cyrix offering.
These 486 wannabe's gained some traction among budget minded enthusiasts, however they failed
to enter the OEM market on any significant scale, although the SLC chip did gain some
favour with laptop manufacturers due to it's low power consumption and ease of installation.
It effectively allowed suppliers to boast laptops with a "486" named chip, at a fraction
of their competition's cost.
Cyrix went on to release the 486SRX2 and DRX2; essentially clock doubled versions of their
They also eliminated some issues their previous chips had with cache control lines on early
It would be May 1993 where Cyrix would start to enter the mainstream, alongside AMD and
Intel, with a Intel 486SX pin compatible CPU, the Cx486S processor, codenamed M5.
This was followed by the DX and DX2 versions in the same year which offered clock doubling,
the same as Intel's DX2, but at a significantly cheaper price.
If only enthusiasts had heard of Cyrix up until now, this would be the point that the
average PC buyer might start to take notice and factor this lower cost alternative into
their decision making process.
Breaking rank from their previous designs, the new processors did benchmark slightly
lower than AMD and Intel counterparts, but their ability to run at 5v compared to the
3.3v of AMD chips meant they could be used as upgrades in early 5v 486 motherboards.
This time round various OEMs took notice, and proudly advertised their new wallet friendly
486 machines, although large brands like Acer and Compaq were wooed only by the AMD alternative,
leaving Cyrix still as a bit of an underdog in this silicon race.
It was around this time that a tombstone was erected in the Cyrix main office stating RIP
Intel Inside, something Jerry was very proud of.
The problem was, 1993 was also the year Intel's P5 architecture landed, with the original
Cyrix, and indeed, other manufacturers such as AMD, had some catching up to do.
This is also around the time of a number of lawsuits, one of many the company seemingly
found itself either instigating or caught up in, spending more each year on suit than
it's original $4 million start up capital.
At the time Intel seemed to have a gripe with every rival x86 manufacturer and gave no exceptions
However, with Cyrix's from the ground up approach, Intel were set to invariably lose.
An out of court settlement then gave Cyrix the right to produce it's own x86 designs
in any foundry which held an Intel licence.
Texas Instruments held a licence, but during 1994, Cyrix would fall out with them over
Instead, an alliance with an alternative licence holder; IBM Microelectronics was established.
Being unable to produce the chips themselves, having a reliable fabricator was essential
for Cyrix and therefore the possibly less than favourable deal struck meant that IBM
earned the right to build and sell half of the Cyrix designed processors under the IBM
This might sound quite dreary, but 1994 would still prove to be a record year for growth
of the business.
By 1995, Cyrix's Pentium clone was still not ready, but instead they had their old trick
up their sleeve.
This trick would be presented in the form of the Cyrix Cx5x86 processor.
Just like their early 386 upgrades, this chip plugged directly into a 3.3v 486 socket and
ran at 80, 100 or 120MHz speeds.
Running at 100MHz yielded performance comparable to a Pentium 75MHz, and this was by far the
most common iteration of the chip..... and so, just like that, us 486 owners seemingly
had an upgrade route to Pentium bliss, but without having to do much more than shell
out a few quid... of course, it wasn't quite that simple.
You had to have a compatible BIOS to begin with, and then of course, there was capabilities.
This time around, Intel had beaten Cyrix to it, and had launched the 486 pin compatible
Pentium Overdrive in February 1995; some 6 months before the 5x86.
However priced at $299, or around £250 over in the UK, this was definitely not a cheap
upgrade route, and what's more, chips like the 5x86 outperformed it in most benchmarks.
Although the Overdrive was far more compatible with the Pentium instruction set, whilst the
5x86 was more like a souped up 486 processor.
Although the 5x86 was affordable, one area of contention was the lack of branch prediction;
a key element of the Pentium line, and I say contention here because, the 5x86 did have
branch-prediction, and indeed the benchmarks Cyrix pushed were with the branch-prediction
features turned on, but due to instabilities, it was disabled before release.
You could in fact, re-enable this through specialist software, but it resulted in an
unstable system, so was fairly useless.
For people like myself who spent weeks pouring over benchmarks seeking an upgrade route,
this was quite a disappointment.
Of course, the processor was much faster than my 486, but it just failed to present the
same optimisations in newer games, or even run them at all, given most software was then
being designed with a true Pentium architecture in mind.
But still, having a 5x86 processor installed in my system always gave me a warm feeling
every time I booted it up.
For some reason I've always enjoyed seeing systems pushed to their absolute limits, and
so for me, this was therefore the perfect processor, whether I'd intended it to be,
Searching through coverdisks for Pentium only software that worked on my 5x86 was my kind
of mustard, but when it did happen I had the smug satisfaction that my CPU was truly ahead
of AMD's Am5x86, which was merely a clock quadrupled 486, trying to shuffle in on the
Although, to be fair, I don't think I ever did find any Pentium only software that would
run on it.
But that would all change.
The 5x86 only had a production run of a measly 6 months as Cyrix diverted all their energy
and production to focus on their new 6x86 processor (codenamed, the M1).
IBM would be left manufacturing their own 5x86 iterations until 1998, but of course,
would also gain rights for the 6x86 under their production agreement.
This sixth generation processor (in the eyes of Cyrix at least) was one of the first true
Pentium rivals on the market, launching on October 6th 1995 and beating AMD's K5 by several
This was their first P5 socket compatible processor and early reviews were very good.
You can see in this PC Magazine Winstone benchmark from March 1996, the Cyrix 6x86 100 is beating
both Pentium 120 and Pentium 133 systems, and for a cheaper price.
There was little not to like, and this would see Cyrix quickly escalate from budget upgrade
path provider, to technological leader of the pack.
The first 3.5v 100Mhz units did need substantial cooling (by 90s standards at least), as can
be seen here, but this was rectified with subsequent 2.8v models.
Of course, Cyrix knew they'd cracked something big here, and in January 96, alongside their
competitors AMD, SGS-Thompson and also IBM had defined the P-Rating system that would
start appearing after March on the official names of their processor.
The 6x86/100, would become the 6x86 P120+, the 110 would become the P133+ and so on.
Based on Winstone 96 results conducted by Microstone Resources, the change was primarily
implemented so that consumers weren't put off by the lower MHz numbers, but it also
made more sense from a comparison perspective.
Some people were baffled & even enraged by the system, but pushed by Cyrix and their
partners, it quickly became the accepted norm.
Of course, if you look at the super-pipelined architecture of the 6x86, it's easy to see
why it outperformed the P5.
It's actually more similar to a Pentium Pro than a bog standard Pentium, with features
such as register renaming, multibranch prediction and speculative execution, and being optimised
for 16 and 32 bit Windows applications, it's performance was bound to excel in the consumer
This really peaked in June with the 6x86 P-200, becoming the fasted x86 based CPU in the world,
and the first non-intel chip to achieve that status in 18 years.
It needed a 75MHz bus, but for those interested in performance, it was an essential upgrade.
Intel may have had superior fabrication abilities with the Pentium manufactured to a .35 micron
process, compared to IBM's .5, but they were falling behind the efficiencies and optimisations
designed into the underdog's chip.
By this point, Intel, were understandably irked, and with little direction to move,
threatened action against the P-Rating system, as people were mistakenly referring to it
as "Pentium Rating" rather than "Performance Rating".
Cyrix quickly resolved this by simply changing the name to PR.
They then went on a rampage about other companies free-loading off their advertising and demanded
all badges be removed from PC cases, apart from the Intel badge.
They might have blown a fuse if it weren't for a savior in the form of Quake.
The 6x86 was designed with business software in mind as a baseline.
Of course it was great at running games as well, but this was primarily a 2D game world.
Quake launching in June, would however, change that.
Whilst developing Quake, John Carmack realised that FPU and integer operations used different
parts of the Pentium core and could be effectively overlapped.
This nearly doubled the speed of FPU intensive game code.
However on Cyrix processors, the operations didn't overlap, effectively halving execution
This is why when I upgraded from a 5x86 to 6x86, pretty much exclusively to play games
like Quake, I was yet again faced with a degree of disappointment.
Quake was evidently, extremely popular and as more and more people used it in their performance
tests, the lofty position held by the Cyrix processors began to slip further and further.
The public tended to simplify the problem, and jumped the 6x86 for having poor Floating
This is a claim Cyrix tried to rebuke through a series of benchmarks on their website, somewhat
at odds with magazine tests, although you'll notice that their 3D Descent test made use
of an accelerator card, and obviously there's no Quake test to be found here.
What I do find amusing is this little paragraph.
"The 6x86 processor performs equal to the comparable Pentium in some cases, while the
Pentium performs better in others.
However, in all cases, both the 6x86 and Pentium processors achieve the smooth motion frame
rate of greater than 13 frames/second"
13 frames per second!
Can you image gamers today getting 13 frames a second and being happy with it??
I hear people complain that games are barely playable when frame rates drop under 60 today.
With multimedia on the rise, Intel would launch the Pentium MMX in October 96 and begin to
thwart the short lived reign of Cyrix.
It's almost as if Jerry Rogers wanted to quit whilst he was ahead, as on December 9th 1996,
he stepped down from his role as President and CEO, remaining on the board of directors.
Co-founder, Tom Brightman would remain as vice president.
Jaw Swent, the Sr. Vice President of Finance and Administration stepped up to chair in
his place, and it's probably no accident he was the president of finance and administration,
because despite their success, Cyrix had posted a net loss of $6.9 million in the year ending
Strangely, they had been making profits of $6 million a year earlier.
This was attributed to increasing competition and the need for substantial price cuts.
But Cyrix were far from over at this point.
Their 6x86 was still a strong contender, and by the end of May 1997, their MMX rivaling
product, the 6x86MX was unveiled.
Remaining socket compatible with it's predecessor and claiming to deliver Pentium 2 performance
at half the price, a new range of P6 processors, Intel had only introduced a few weeks beforehand.
They had also begun a 2 prong attack by launching the new MediaGX processor in February, aiming
to create a whole new budget PC market.
Based around their existing 5x86 chips, these new processors offered an all in one PC solution
and was in fact one of the first PC's on a chip (well, there was also a Cx5510 companion
chip to handle sound and some of the bus functions, but almost).
Integrating their XpressRAM, XpressGRAPHICS and XpressAUDIO all within the same processing
At just $99 for the 133MHz flavour compared to say $407 for a 166MHz Pentium MMX, without
the additional components, this really did open up a new landscape and is essentially
the main harbinger for the true budget low end PC market which Intel and AMD would need
to catch up with.
By now, Cyrix had no troubles getting OEM manufacturers on board.
Compaq were one of the first with their Presario 2100 storming in at under $1,000, which seemed
to create all sorts of hype.
Sure, it wasn't going to knock the socks off a P200 owner, but here was a kinda, almost
Pentium with 24 megs of RAM, 2.1GB Hard Drive, 8X CD-ROM drive and soundblaster compatible
sound for a reasonable price.
I mean, you had to pay extra for the monitor and savvy consumers could pick up a lesser
known brand, with a better spec, for about the same price, but with the Compaq badge,
it caught the media attention, and that's exactly what Cyrix needed.
It wasn't long before other big name manufacturers started shipping Cyrix processors in their
machines, seemingly moving their brand image up a notch.
Steve Tobak, who was Vice president of marketing for Cyrix at the time, has written about how
he thought Tom was nuts for dreaming up the sub $1000 PC concept, but after thinking about
PC price elasticity, describes how a big light bulb went on.
It seems consumers were on the same train of though.**
After three consecutive quarters of losses, Cyrix posted a profit for the first-quarter
of 1997, but their previous posted revenues of $183.8 million were still a tiny dent in
the $15.4 billion worldwide processor market.
Still seeking a CEO, and looking favourable, Cyrix seemed ripe for a buy-out, and National
Semi-Conductor seemed to agree.
Aquiring Cyrix in November 1997 for a $550 million stock exchange, as an autonomous and
wholly owned subsidiary.
It was the MediaGX which had really roused their interests, with Brian Halla, Semiconductor's
CEO stating "we intend for the Cyrix acquisition to give us the building blocks to provide
system-on-a-chip soltuions for sub $500 PCs and a broad range of low-cost information
Seemingly advancing the 6x86MX design to keep pace with Intel and AMD was not on their agenda,
however Cyrix would later announce that it's merger would not change their development
or marketing plans.
This focus on the MediaGX did create sales, but with less focus on staying with the competition,
it wasn't long before both Intel and AMD were outpacing Cyrix once again.
The new "Slot one" Pentium 2 had launched in May 1997, leaving other manufacturers stuck
in Socket 7 motherboards.
Rival AMD followed up with their impressive K6 processor range in April.
Interestingly, the Pentium 2, and pro designs actually created a legal dispute whereby,
this time, Cyrix accused Intel of copying their power management and register renaming
methods for use in the Pentium 2 architecture, but this was settled quite quickly after the
National Semiconductor buyout, with a cross licence agreement, allowing both parties to
borrow each others patents.
Although not stipulated, this suggests somewhat that Cyrix had a good case for this, but was
undoubtedly eased by National Semiconductor already holding a licence for Intel's chip
Unfortunately, the world began to look bleak for Cyrix.
Before 1997 was out, they had to issue a 10,000 chip recall on their 6x86MX P200 line due
to high failure rates.
This may have been partly to blame for the delay on their Pentium 2 rival, the M2, released
in early 1998.
The chip may have held the number 2 in it's name, but really, it was just a re-badged
6x86MX available in some higher clocked iterations, but all of which were really pushing their
manufacture process at the time.
The floating point and integer capabilities were now drastically falling behind Intel's
and AMD's new processors, and coupled with this, Intel had released budget Celeron versions
of their Pentium 2 processors in a bid to reclaim the sub $1000 PC market, National
Semiconductor were keen on seizing.
By September 98, the AMD K6-2, with it's superior performance, accounted for a 68% of sub $1000
PCs; a reported cache bug in systems with more than 32MB of RAM was no issue at this
level, whilst Intel's Celeron was level pegging the MII at 16% each.
Technology vs. cost was accelerating at such a rate, that the MediaGX, even having reached
a 266MHz MX iteration, was already becoming obsolete, especially with whisperings of a
500MHz AMD K7 on the horizon.
National Semiconductor would end their production arrangement with IBM at the end of the year,
having brought most production in house at their new manufacturing facility in Maine
and having already launched 300 and 333 iterations of it's M2 on a 0.25 micron process.
Their next step was to try and rival their competition with a 0.18 micron process the
following year with the MII 433 (which ran at 300MHz), and then release a range of new
Cyrix knew they were falling behind, acknowledging poor performance under certain games with
their MII, they seemed keen to retain their budget market whilst also catching up with
the pack leaders.
Their Senior Vice President, Kevin McDonough was quick to stave off fears by throwing a
range of names about.
In development were 2 cores, the Jalapeno and the Cayenne.
The Cayenne was designed for 3 variations, the first code-named, Jedi was for Socket
7, labelled to be their next offering and looking to incorporate the 3DNow! extensions
along with better FPU performance.
The second was Gobi, a socket 370 compatible CPU that would push the technology further,
and the third was the MXi, offering integrated graphics and designed to replace the MediaGX.
The Jalapeno core was their true next generation, pegged for release in the Mojave processor
after the end of the millennium, which I wanted to say because it sounds epic.
The Jalapeno would have a dual issue FPU, an 11 stage pipeline and integrated 3D graphics
that could store textures in it's L2 cache, and by the sound of things, really could have
put Cyrix back among the pack leaders.
In this late 1998 interview, McDonough seems very upbeat at the company position, including
putting their MediaGX to use into all sorts of products, such as set top boxes and fridges.
He even seemingly states that they could be in a position to completely throw Intel off
track, but thankfully weren't looking to put them out of business.
However, this, through circumstance really became a case of all mouth and no trousers,
mainly because the trousers didn't have any money in their pockets.
By 1999, AMD and Intel were leap-frogging each other with clock speeds, and not only
Cyrix, but National Semiconductor were finding themselves in financial difficulty.
Their intentions and plan seemed credible, but unfortunately, they'd fallen too far behind.
On June 30th 1999, Via Technologies would acquire the Cyrix chip designing devision
for the net sum of $167 million, followed almost immediately by Centaur Technology from
Integrated Device Technology for $51 million, allowing Via to enter the x86 processor market.
Via, like Cyrix started out, was a fabless company, and so manufacture would continue
at National Semiconductor's plant in Maine.
There are some stories from Cyrix employees knocking around from this era, and it paints
a rather interesting picture of events immediately prior to acquisition;
After talking to the engineers at their Richardson base, VIA's CEO, Chen Wen-Chi was a little
surprised to hear some projects were behind in their design schedules, somewhat at odds
with McDonough's interview the year prior, so an agreement was made for National Semiconductor
to lay off half of the Cyrix staff at their cost.
Via's interest was really in the Cyrix brand and their IP, which as we know, had a separate
Genus to Intel's processors, and could therefore be used in defence against an Intel lawsuit
aimed at Via's current chipsets and any future processors.
Cyrix's R&D costs were at that point, $10 million per month.
The Mojave concept was very far off and so was immediately dropped along with all other
projects, apart from Gobi, which was almost at the sampling phase.
This was the point when Tom Brightman would leave the company, and between a lot of good,
experienced staff being let go, and others simply being snapped up by competitors, Via
was left with Cyrix's back catalogue, their brand, Intellectual property and the Gobi
project & staff.
The Gobi code-name was then renamed to Joshua, keeping in line with Chen Wen-Chi’s, biblical
Joshua then, poised to become the Cyrix M3, with its 7-stage Cayenne core was expected
to top out at around 600MHz, but was consuming some 25 watts.
The Centaur Technology team Via had also snapped up, had also been working on a WinChip core,
which was by now known as Samuel.
This core had 12 stages, allowing for more effective overclocking, at only drew 6w at
similar clock speeds.
The decision was therefore taken by Via to stop production of Joshua and go full speed
ahead with Samuel.
This is the processor that would ultimately be branded the Via Cyrix III and launched
This was seen by many as a poor move, given the potentially superior abilities of the
Joshua chip, that were speculated to outperform the AMD K7, but on a much smaller die.
The C3 may have not been the most powerful package available, but it did beat the competition
in power consumption, and met the demands of the embedded marketplace quite well, finding
use in notebooks, with the later C7 revision finding it's way into the HP 2133 Mini-Note
The last processor to retain the "C" part of the Cyrix name, the Via C7-M released in
2007 as a mobile processor.
But this really leaves the MII 433, in April 1999, the real last Cyrix processor.
But that's not the end of the story.
Although VIA bought out most of Cyrix, National Semiconductor actually held onto the MediaGX
It seems they really did want to hold onto the chip that had forged their initial deal
back in 1997.
National Semiconductor renamed the MediaGX to The Geode at the end of 1999 and sold them
mainly for use in thin clients and industrial control systems.
These processors were developed further to the Geode GX2 in 2002 before being bought
up by AMD and sold until 2009 under various guises and iterations.
So, for those hoping to cling onto the Cyrix dream; there very well might be a few strands
of Cyrix left in AMD processors here and there, but 2009 pretty much draws the end of the
line for the work Cyrix began way back in 1987.
The Cyrix story is one hell of a ride, but I suppose their biggest accomplishment was
laying the foundations for the budget PC market, without really sacrificing performance, and
they did it all without the funds other companies held.
Indeed, that alone is one of the reasons I have so much passion for the Cyrix brand.
Between 1995 and 1998, the Cyrix brand allowed me to upgrade my PC to levels I simply wouldn't
have enjoyed without their presence in the market, and sure, sometimes their CPUs didn't
live up to expectations, but for me, that just made it all the more fun.
Everyone loves an underdog, right?
In a way, these processor rivalries of the 90s, felt very much like the Spectrum vs.
Commodore, Amiga vs. ST and Sega vs. Nintendo battles of years gone by.
It didn't really matter if your hardware was technically inferior or not.
It didn't matter because it was yours, it brought you joy, and for that reason it was
worth defending to the very end.
Things haven't really changed today, have they?
I tell you what, as you've stayed till the end, here's a bit of Cyrix trivia for you;
The film Eraser featured a defense corporation known as "Cyrex".
Cyrix got a bit upset about the potential name confusion, so asked for it to be changed.
At a cost of some $10 million and over the course of a week, the name was then retroactively
edited to become "Cyrez", both digitally, and vocally, with some actors having to re-record
You can see some remnants of this in this scene, with the Computer code, clearly an
abbreviation of Cyrex rather than Cyrez.
Thanks for watching.
Contribute, or watch something else.
In any case, have a great evening!