• Please review our updated Terms and Rules here

Anyone got love for Cyrix?

EvanW

New Member
Joined
Dec 2, 2021
Messages
1
I didn't see a place that collectors of Cyrix might gather on here.
Right now I have a small collection of roughly 100 machines (many work some need repair) everything from some early 286 systems to the final 6x86 and Cyrix MII.

I am actively working on collecting historical documents and would like to see if I am alone or if there are compatriots out there.
 
I didn't see a place that collectors of Cyrix might gather on here.
Right now I have a small collection of roughly 100 machines (many work some need repair) everything from some early 286 systems to the final 6x86 and Cyrix MII.

I am actively working on collecting historical documents and would like to see if I am alone or if there are compatriots out there.

Welcome aboard Evan. There are plenty of Cyrix lovers here. How about posting some pictures of your hoard.
 
Funny that this is mentioned today. Needing to run NT 4.0 terminal server for a test, I dug up an old PR300 system with a CF card as hard disk. I was surprised how quickly the old system booted, given the hardware (Amptron 8600D).
 
I was into Cyrix back when they were a major maker of FPU for 386 machines and my first Pentium class 166+ CPU was Cyrix (which ran too hot and was replaced a slower Pentium 133).

Amptron was another brand of PC Chips motherboards I believe.
 
My Cyrix system is still listed in the BogoMIPS mini-howto for Linux ( https://tldp.org/HOWTO/html_single/BogoMips/ ):


6x86/120 P120+overcl 104.86 Howard Poe <falcor@_kingsnet.com>

It was an unusually high rating due to the way this particular motherboard cache could hold the whole algorithm, note it's double a similar system on the list. This was back when the Cyrix P120+ was new.
 
I was a big fan of the 6x86 because the performance was a lot better than an equivalent Pentium if you were a Luddite like me and still mostly running 16-bit integer code.
 
First "pentium-class" machine was a 6x86-166 Cyrix. My real intro to Win95 and WinNT 4 at home. Still have it.... somewhere.
 
My first “Pentium-class” machine was an AMD K5-90, but I suppose I have fonder memories of the kind of rotgut Cyrix P166+ I threw together about a year later. The case I used was basically made of tinfoil, but the motherboard I snagged off the back of a truck had 64MB of SDRAM instead of EDO and felt *fast* for the time. (My benchmark for comparison was the 233mhz Pentium II at the office.) The chip was the later IBM manufactured one so I never ran into the heat issues of the early chips.

Of course the Achilles heel of the thing was floating point, the K5 might have been faster at Quake timedemos. But I do think that weakness was highly overblown at the time, for most software it ran great.
 
final 6x86 and Cyrix MII.

Cyrix's designs lived on long after the MII. The NSC and later AMD Geode GX was a Cyrix 5x86 core die shrunk and made into a SoC. AMD used them for various projects, one being their failed 50x15 PICs that ran WIndows CE. You could hack them and get an unlocked BIOS on them to load other operating systems though. I have one that I hacked around a decade ago, it makes a screaming fast Windows 98 and DOS box in a tiny form factor. 400 MHz 5x86 with 512 MB of RAM and a 10 GB HDD.

The chip was the later IBM manufactured one so I never ran into the heat issues of the early chips.

While Cyrix CPUs did have higher than average power consumption and heat output, that wasn't the big issue. The big issue were their CPUs that used the weird 83 MHz bus speed that heavily overclocked system components and ran them so far out of spec that it caused severe stability issues. PCI would run at 41.5 MHz, AGP would usually run at the FSB so 83 MHz and memory would also be heavily overclocked until PC-100 became available. Their reputation was tarnished on their 83 MHz FSB CPUs that made unreliable machines. They did eventually tone it back down to 66/75 MHz and later I think a couple of 100 MHz FSB parts, but it was too late by then.

Of course the Achilles heel of the thing was floating point, the K5 might have been faster at Quake timedemos.

Something else that crippled potential performance was bugs in the core that caused features to have to be disabled. Cyrix top brass promised things like branch prediction that were disabled in final silicon because of severe bugs that made the CPU unstable. There were a bunch of CPU features that weren't enabled on final silicon that greatly impacted performance. Vogons has a whole thread discussing it and the software that can be used to re-enable branch prediction and other features on the fly. Cyrix apparently fixed many of the bugs in later CPU mask revisions and some of the features were able to be manually enabled without or will less stability issues.
 
While Cyrix CPUs did have higher than average power consumption and heat output, that wasn't the big issue. The big issue were their CPUs that used the weird 83 MHz bus speed that heavily overclocked system components and ran them so far out of spec that it caused severe stability issues. PCI would run at 41.5 MHz, AGP would usually run at the FSB so 83 MHz and memory would also be heavily overclocked until PC-100 became available. Their reputation was tarnished on their 83 MHz FSB CPUs that made .

The 166+ was a 66mhz bus so it didn’t have that problem. The heat issues I’m referring to are with the original (non-6x86L) single voltage version of the M1. Those drew close to half again the wattage of a Pentium and as a result would sometimes take out VRMs, which gave the chip an initial black eye. The 2.8v core “L” versions were fine.

The weird bus speeds for the MII were a seriously bad idea, at least with overclocked Intel or rotgut chipset motherboards. My vague memory is that some third party chipsets could do a 2.5 multiplier for PCI, but if you were going to blow the money to do it right the full Super Socket 7 100mhz bus K6-2 suddenly made way more sense.
 
The 166+ was a 66mhz bus so it didn’t have that problem. The heat issues I’m referring to are with the original (non-6x86L) single voltage version of the M1. Those drew close to half again the wattage of a Pentium and as a result would sometimes take out VRMs, which gave the chip an initial black eye. The 2.8v core “L” versions were fine.

That can't entirely be blamed on Cyrix, they just exposed a design deficiency in CPU power regulation on motherboards of the time. By the late 486 era, motherboards should have been using switching regulation for the CPU vcore. As the 486 got past 66 MHz, 5v for the core was producing unacceptable levels of heat, which lead to later 486 parts having a vcore of 3.45v and lower.

In the Socket 5/7 era when vcore started to drop lower, motherboard manufacturers were still using linear regulators for vcore. You had 3.3v and lower running off the 5v rail, which was at the near limit of the linear regulators being used, pushed over the edge by using grossly inadequate heatsinks. Necroware on Youtube has a video going over replacing the linear regulator with a switching regulator on an old Socket 7 board, and even adding in support for CPUs never designed to be used on the board.

https://www.youtube.com/watch?v=CMiGVQbMC5U

The weird bus speeds for the MII were a seriously bad idea, at least with overclocked Intel or rotgut chipset motherboards. My vague memory is that some third party chipsets could do a 2.5 multiplier for PCI, but if you were going to blow the money to do it right the full Super Socket 7 100mhz bus K6-2 suddenly made way more sense.

Yeah, I can't really see why Cyrix was doing the 83 MHz bus; Other than maybe because they couldn't get their CPUs to clock ramp like AMD and Intel, they tried to make up for it in a faster bus. 75 MHz was fine, because you could keep things mostly in spec, but 83 MHz was not at all.
 
I do recall Cyrix doing the odd 83mhz bus, same thing Apple did with the G3 Wallstreet and then ditched on the Wallstreet 2 Revision back to 66mhz for some reason.
 
The Beige and B&W G3 had a setting for an 83 MHz bus, but it had the same problems as PCs did with running everything out of spec. I don't think Apple ever officially released any desktop model G3s with anything but a 66 or 100 MHz bus.

I remember years ago when I was installing my Sonnet G4 upgrade, I was reading about the jumper block settings to get it right and found a page which is probably lost to time now. It described a guy testing various clock multipliers and FSB speeds, and 83 MHz was never stable due to the memory and PCI bus running out of spec. Later PCI 2.1 cards which could tolerate a 66 MHz bus would have probably been fine, but not too many of those made it over to the Macintosh, due to requiring a PowerPC ROM to work, if the card needed its own option ROM.
 
I've been playing with a 386 motherboard I pulled from a DTS head unit, it took a bit to get working properly because of battery bukakke. I just got the motherboard cache working, when I discovered the CPU had internal cache too. It's IBM's version of the Cyrix 486SLC2, the clock doubled version of Cyrix's weird 486 in a 386 package. Cyrix had either 1 or 8 kb of cache, but IBM doubled that to 16 kb of L1 cache.

So I have a 386 motherboard with a 486SLC2 @ 66 MHz with 16 kb L1 cache and 64 KB L2 cache.

Nk8lgktl.jpg


All of the cache makes it a pretty fast motherboard. Speedsys says it's around a 486 DX/2 50 and it feels about that fast.
 
I don't think IBM's 486SLC had anything to do with the Cyrix/TI 486SLC. They were independent developments. IBM previously made the 386SLC "Super Little Chip" with 8K internal cache. The 486SLC increased the cache to 16K, added the 486 instruction set, and introduced clock-doubled and clock-tripled versions. I had a 486SLC2-66 motherboard with the tiniest CPU cooling fan I've ever seen -- one inch square!
 
I don't think IBM's 486SLC had anything to do with the Cyrix/TI 486SLC. They were independent developments. IBM previously made the 386SLC "Super Little Chip" with 8K internal cache. The 486SLC increased the cache to 16K, added the 486 instruction set, and introduced clock-doubled and clock-tripled versions. I had a 486SLC2-66 motherboard with the tiniest CPU cooling fan I've ever seen -- one inch square!

I remember setting up a 486SLC machine in '92 with DOS 5. I didn't pay attention to details that closely in those days, but I remembered it was just a 'slow' 486 that was really a variant of the 386. I don't remember any VLB on it. Now I know that to be true, likely because of the 16-bit bus limitation. But it is quite interesting it had a 16K L1, so maybe not too bad if it was a budget chip if you didn't really need 32-bit I/O in windows or FPU.

Of course, Cyrix's big days were to come a bit later... we were fans because their chips basically dropped in and made things quite a bit faster without needing to completely change out the board for an expensive intel chip.
 
Last edited:
I upgraded my Compaq 386/20e with a Cyrix 486DRx2 and FasMath. It definitely got a boost in performance and capability. With its full compliment of 16MB and larger/faster HD, it can run Windows NT 4.0 acceptably which it wouldn't have been able to do without the upgrade.
 
I don't think IBM's 486SLC had anything to do with the Cyrix/TI 486SLC. They were independent developments. IBM previously made the 386SLC "Super Little Chip" with 8K internal cache. The 486SLC increased the cache to 16K, added the 486 instruction set, and introduced clock-doubled and clock-tripled versions. I had a 486SLC2-66 motherboard with the tiniest CPU cooling fan I've ever seen -- one inch square!

The nomenclature IBM and Cyrix used around that time is indeed incredibly confusing. The 486SLC2 is, yeah, IBM's core, not Cyrix's, and is a pretty amazing piece of work considering it could score right up alongside a "real" 486 in benchmarks despite being trapped in a 386SX's body, 16 bit bus and all. But Cyrix had their own "Cx486SLC" which was essentially the same concept (But a much smaller cache, only 1K) that mostly ended up in laptops What really can give you a headache is IBM's "Blue Lighting" label, which they themselves slapped on two completely different chips. There was a 32 bit bus version of the SLC2 (IBM core) that was sold as both the "486BLx" and "486DLCx" (but only welded to motherboards, IBM's deal with Intel didn't allow them to sell socketed chips), with the latter designation likewise conflicting with the Cyrix Cx486DLCx/Cx486DRx line (386 socket). But apparently that wasn't confusing enough because later IBM *also* slapped the "Blue Lightning" label on some Socket 3 products, but those *were* licensed versions of Cyrix's actual Cx486DX, not the IBM core. So one Blue Lighting isn't necessarily the same as another Blue Lightning either.

There was this brief moment in the 1990s where they were selling the *ell out of IBM SLC/2 motherboards and generic-feeling small form-factor systems in the back pages of Computer Shopper. I remember at the time being impressed how well they ran considering the limitations, but also being a little confused why IBM was pushing a chip limited to 16MB of RAM at the same time they were trying to make OS/2 2.x relevant. They were still selling these right up to around the time Windows 95 came out, I think. Buying one in 1995 would definitely qualify as a pretty serious mistake.
 
Back
Top