Today I got a Cyrix 100GP bought "as is" and decided to test it on an M919.
I mistakenly set the multiplier to 4x. I didn't find out until I checked the CPU frequency.
It was indeed 4x33=133MHz.
How can this be? It's the first time for me to see a 100GP have a 4x multiplier.
And it does 133MHz with the standard 3.45v setting without any problems so far. I performed basic stability tests such as Quake, and it's stable. Enabled some registers from a .bat file I prepared and yeah, it's fast for a 486, especially with a 33MHz bus. Formerly I run an Am5x86@180MHz (3x60) and it was fast too. But this thing's FPU is far superior.
So back to the main question, has anyone else seen one of these??
If the CPU has 100GP printed on the surface, but does not contain the /4x, it will likely fail at 133 MHz in Widows 95. Sometimes they fail immediately at boot, other times they fail 10 minutes later.
From my best guess, the units with "100GP/4x" were failed 133 MHz models and marketed for 25 MHz bus systems, normally in upgrade interposers. Similarly, 120GP/4x usually fail at 133 MHz. I've tested one 120/4x that failed immediately, and another 120/4x that would be considered stable for most retro purposes.
It also seems that Evergreen was shipped some Cyrix 5x86 chips with no speed markings, presumably leaving it up to Evergreen to mark their speeds on the upgrade box. I have had the best luck with unmarked at 133 MHz. I use one such CPU in my Cyrix 5x86-133/4x system.
With any luck, you have a 100GP /4x chip which is stable at 133 Mhz for most purposes, however I'd suspect it would need 3.6 - 3.75 V to get close Windows stable.
Attach a photo, top/bottom, of what you have.
Plan your life wisely, you'll be dead before you know it.
I've been testing it for hours.
The only issue that I noticed is that even if I run the CPU @100MHz without any register tweaks, my version of Vim (v5.4 32-bit) makes the computer soft crash, i.e. it runs and exits normally but after that, most programs won't start and the system crashes.
Everything else has been stable for hours so I suppose that's something about this particular Vim version and the Cyrix CPUs, or maybe the motherboard.
So far this 100GP turned out to be better than my two 120GP.
Here's some pictures:
The attachment cyrix.5x86.100gp.jpg is no longer available
The attachment bottom.sticker.jpg is no longer available
Overall only the FPU performance is noticeably better than an Am5x86 @180MHz.
14.6% better FPU performance, and 2.44% better CPU performance on Dr.Hardware v9.0e.
The attachment drh9.png is no longer available
Here's also a period correct Dr.Hardware 3.1
The attachment drh31.png is no longer available
Speedsys
The attachment speedsys.png is no longer available
Despite that, Quake gives me 17.2 FPS vs. 20.44 FPS with an Am5x86 @180MHz. I suppose that's related to the overclocked PCI bus.
Is it possible to boot with a 2x mutliplier? That would be interesting.
With any luck, you have a 100GP /4x chip which is stable at 133 Mhz for most purposes, however I'd suspect it would need 3.6 - 3.75 V to get close Windows stable.
The CPU is likely getting something close to that. The C3420 transistor outputs 3.653v.
Feipoa can you recommend any stability test? For all my tests I used the register settings you suggest on this thread :
What's Vim? I only know Vim as a bathroom cleaner.
What motherboard are you using? Not all motherboards support all the register enhancements of this CPU, particularly LINBRST and BWRT.
Your CPU was from week 7 of 1996, the last week of production. They were pushing for 133 Mhz this week. Every CPu with a 100GP stamp that I've tested wasn't fully stable at 133 Mhz.
The Cyrix 5x86's FPU tends to do well in synthetic benchmarks compared to Am5x86 180, but the real tests are the games, as you've discovered with Quake.
It wouldn't be easy to boot your 4x chip at 2x. You are better off using a 3x chip set to 2x mode if you want to run 2x66 Mhz, but only two boards are known to support this configuration - the 8433UUD and LSD.
Interesting that you are running at 3.65V. Hopefully you are stable at 133 Mhz and I have a gap in my test base.
For stability, I recommend Windows 9x and NT4, in particular playing around with IE5/6 and loading google and 68k.news. If that goes well, try running an mp3 128kbps while doing IE5/6. Then keep agitating it more from there, like opening PDF files, GLQuake, etc.
BTB won't be stable in Windows on your CPU. You'd need S1R3 for that.
Plan your life wisely, you'll be dead before you know it.
I'm glad to announce that fortunately this chip is completely stable @133MHz 3.65v
It's been working for many hours under heavy load with the modified registers for S0R5 (win).
What motherboard are you using? Not all motherboards support all the register enhancements of this CPU, particularly LINBRST and BWRT.
It's the M919. It took me quite some time to do all the tests because of the quirky nature of the M919 and because I also tested other Cyrix and Am5x86 CPUs to compare the performance.
1. It was very important to enable 'Linear Burst for Cx5x86' in the BIOS, otherwise the Peter Moss' utility crashes the system.
2. The EDO ram only serves one purpose: to use 60MHz 0/0 W.S. It doesn't offer a benefit over FPM at the same speed, but I couldn't find FPM modules that support 60MHz 0/0 W.S. The ones I have need 1/0 W.S to be stable.
3. Installing the RAM module on the slot closer to the edge of the board improves stability @60MHz.
Only one of my Am5x86 runs well @180MHz 4v, it's an ADW A9630 BPG.
My best compatible video card is a Radeon 9250 PCI (no Nvidia or Voodoo unfortuntely). The earliest compatible drivers are version 3.6. The max this card can do is 60MHz, it won't post at 66MHz.
The attachment sandra.m.png is no longer available
Overall the Am5x86 feels much faster @180 or @160. It feels closer to a Pentium experience. According to most synthetic benchmarks the ALU is close to a Pentium 100, and the FPU to a Pentium 66. The desktop experience is better, the system is more responsive. RTS games such as Starcraft become more playable, but it's still a bit too slow for such games. I'm running Win98SE. That may or may not hinder performance.
Is it the same to set the registers before starting Windows or during the Windows session?
The BIOS this motherboard has is dated 5/6/96. I've seen a lot of different versions... is there a consensus as to which is the better one?
What's Vim? I only know Vim as a bathroom cleaner.
Vim is one of the most used command line text editors. It has 16-bit and 32-bit DOS versions and they usually work fine. There seems to be a problem with Cyrix processors though.
Was it tested in a case with the lid on? Your experience may be different with the lid on during the summer.
How much memory are you using? M919 v3.4 at 3x60 MHz, I could not get stable at 0/0ws unless using only a single stick of 16 MB EDO.
Did the Radeon 9250 also work with DirectX games in Win9x?
You only set registers before loading Windows, not in Windows.
Yes, M919 won't do 66 MHz FSB.
I don't know what the best version of the M919 BIOS is. There have been reports that the 1998 version contains the 64 MB bug, while the one you are using does not. I've yet to verify this on my system.
Plan your life wisely, you'll be dead before you know it.
taurowrote on 2025-02-09, 13:52:I'm glad to announce that fortunately this chip is completely stable @133MHz 3.65v
It's been working for many hours under heavy […] Show full quote
I'm glad to announce that fortunately this chip is completely stable @133MHz 3.65v
It's been working for many hours under heavy load with the modified registers for S0R5 (win).
What motherboard are you using? Not all motherboards support all the register enhancements of this CPU, particularly LINBRST and BWRT.
It's the M919. It took me quite some time to do all the tests because of the quirky nature of the M919 and because I also tested other Cyrix and Am5x86 CPUs to compare the performance.
1. It was very important to enable 'Linear Burst for Cx5x86' in the BIOS, otherwise the Peter Moss' utility crashes the system.
2. The EDO ram only serves one purpose: to use 60MHz 0/0 W.S. It doesn't offer a benefit over FPM at the same speed, but I couldn't find FPM modules that support 60MHz 0/0 W.S. The ones I have need 1/0 W.S to be stable.
3. Installing the RAM module on the slot closer to the edge of the board improves stability @60MHz.
Only one of my Am5x86 runs well @180MHz 4v, it's an ADW A9630 BPG.
My best compatible video card is a Radeon 9250 PCI (no Nvidia or Voodoo unfortuntely). The earliest compatible drivers are version 3.6. The max this card can do is 60MHz, it won't post at 66MHz.
The attachment sandra.m.png is no longer available
Overall the Am5x86 feels much faster @180 or @160. It feels closer to a Pentium experience. According to most synthetic benchmarks the ALU is close to a Pentium 100, and the FPU to a Pentium 66. The desktop experience is better, the system is more responsive. RTS games such as Starcraft become more playable, but it's still a bit too slow for such games. I'm running Win98SE. That may or may not hinder performance.
Is it the same to set the registers before starting Windows or during the Windows session?
The BIOS this motherboard has is dated 5/6/96. I've seen a lot of different versions... is there a consensus as to which is the better one?
What's Vim? I only know Vim as a bathroom cleaner.
Vim is one of the most used command line text editors. It has 16-bit and 32-bit DOS versions and they usually work fine. There seems to be a problem with Cyrix processors though.
I also run one of these on a 919 3.4 at 2x60mhz 3.7 ish volts. I'm using an IBM marked chip (I have two identical ones, they each behave 100% identically). I am also running one of the mythical 1024kb cache modules.
My memory is a 64mb stick with 50ns chips, I found I needed 1/0 ws. I could run all day long in DOS at 0/0 but 1/0 was needed for stability of some windows games. I will try moving it to the outmost slot like you have, I was myself wondering if this would help help stability by not leaving the outermost slot unterminated so to say. It probably doesn't matter at the bords intended range of FSB speeds of 33 or 40 mhz, but I can see it having some effect at 60.
Are you sure edo isn't faster then FPM? Do you have the bios setting set appropriately?Because I have tested this and it was indeed faster, however I only really run this board at 60mhz so not sure if the difference may be negligible at lower FSB speeds.
I run windows98se. I have run in to a few games that dont like a particular cyrix enhancement - action supercross will eventually crash with membyp enabled, or space cadet pinball will crash with fp_fast enabled. If I go to play any of these I just run a BAT file before the game, withing windows, that runs the peter moss utility again and disables whatever feature I need off. I have not had a problem with this, aside that maybe once in 20 attempts this will freeze the system.
Feipoa is right. This was all rock solid running in the open. Once encased, and especially after adding a matrox g200, I'd get random instability after "heat soak". This problem has seemed to significantly lessened since adding a case fan, something I've never had to do on a 486. Clearly the chipset is running on the edge of it's capabilities and it doesn't take much to make it unstable.
Obviously the g200 and 1024 of cache is a bit of overkill, however, this combination now enables me to play c&c tiberian sun or worms armageddon at 800×600, or outlaws at 640x480 at "acceptable " framerates, something I was not able to do before, and quite frankly, something I'd never really thought possible on a 486.
bertrammatrix:
What BIOS are you running with that single 64 MB module? If you are using any of the BIOSes later than 5/6/96, check your DOS Quake score with the 64 MB chip, then test it again with a 32 MB chip (or even 96 MB total). If your Quake score is substantially lower with 64 MB total, you have run into the "64 MB bug".
I concur, Outlaws is definitely playable and enjoyable on a hyper 486.
Yes, some games don't like particular Cyrix enhancements, but you may also discover that reducing your CPU operating frequency may let some enhancements still work. Most unfortunate are the games which don't like FP_FAST. I think Turok was one such game.
Plan your life wisely, you'll be dead before you know it.
It should be faster, but the chipset or the CPU can't keep up.
If I could I would use a Voodoo3. I only have a Banshee and it's not compatible with this PCI implementation.
I've seen Starcraft run smooth on an ISA card so something must be up with your driver...
We have a different definition of smooth then.
The game is perfectly playable and I'd have been satisfied as a kid in 1998 but comparing it to a Pentium MMX or K6-2 machine, now I can say it's not the best experience.
Was it tested in a case with the lid on? Your experience may be different with the lid on during the summer.
It's sitting on a test bench. I noticed this CPU is very sensitive to temperature so I'm using a beefy 462 heat sink with a slow 120mm fan on top of it to keep it cool. It's not even lukewarm.
How much memory are you using? M919 v3.4 at 3x60 MHz, I could not get stable at 0/0ws unless using only a single stick of 16 MB EDO.
I'm using a single 32 MB EDO stick. I'm also using a PCI Promise IDE controller to improve system performance since the on-board IDE controller is so slow.
I also run one of these on a 919 3.4 at 2x60mhz 3.7 ish volts. I'm using an IBM marked chip (I have two identical ones, they each behave 100% identically). I am also running one of the mythical 1024kb cache modules.
Nice. I only have the original "01S3" 256K module btw.
My memory is a 64mb stick with 50ns chips, I found I needed 1/0 ws. I could run all day long in DOS at 0/0 but 1/0 was needed for stability of some windows games. I will try moving it to the outmost slot like you have, I was myself wondering if this would help help stability by not leaving the outermost slot unterminated so to say. It probably doesn't matter at the bords intended range of FSB speeds of 33 or 40 mhz, but I can see it having some effect at 60.
It did for me! But these PC Chips boards are truly a lottery as someone said many years ago.
Are you sure edo isn't faster then FPM? Do you have the bios setting set appropriately?Because I have tested this and it was indeed faster, however I only really run this board at 60mhz so not sure if the difference may be negligible at lower FSB speeds.
If you use your system at 33MHz for example, there won't be a difference in performance whether you use EDO or FPM. At least that's what I could measure.
I run windows98se. I have run in to a few games that dont like a particular cyrix enhancement - action supercross will eventually crash with membyp enabled, or space cadet pinball will crash with fp_fast enabled. If I go to play any of these I just run a BAT file before the game, withing windows, that runs the peter moss utility again and disables whatever feature I need off. I have not had a problem with this, aside that maybe once in 20 attempts this will freeze the system.
Very interesting. I should try more games but I don't really know which ones to try. I'll try to find Space Cadet.
As feipoa pointed out, the S1R3 accepts BTB and shows an improved ALU performance, but this 133MHz ninja is a S0R5 part.
Obviously the g200 and 1024 of cache is a bit of overkill, however, this combination now enables me to play c&c tiberian sun or worms armageddon at 800×600, or outlaws at 640x480 at "acceptable " framerates, something I was not able to do before, and quite frankly, something I'd never really thought possible on a 486.
Your results with the enhancements enabled, or as to which one exactly may cause issues may vary wildly across hardware, I now fully understand why cyrix/IBM just decided to keep them off by default.
On several LS486 boards I would get occasional freezes during the launch of Road Rash or Worms Armageddon with fp_fast enabled, but never on the 919. But then Across never had a problem with membyp on a m918 (avoid this board if you like your sanity).
You are very lucky if you have cyrix branded chip that's able to run at 133 reliably. I have been through a few but it was always just the IBM'S were long term stable at 120, and of those IBMs I could never get one to run at 133. I could often complete post, and on the m918 I was able to actually run a few dos benchmarks, however once the cpu warmed up it was all over. I assume with a peltier or ice I could do it. I don't think it was a motherboard issue at 66mhz as I was able to use fairly fast timings on the 918 at 60mhz, it seemed wholly a thermal issue.
I am actually in the other boat right now, trying to find an AMD to run at 3x60 to get all I can out of this thing 😀
🤣
I have an M921. It's a very compact and nice board. Very similar to the M919 minus the VLB slot and the fact that it's FPM only. I think the reduced size may help with the innate instability of the 60MHz bus or even the 66MHz bus. I also recently did some tests with this board but things got messy and I gave up. 486 OCing takes SO much time and it's so unpredictable!
You are very lucky if you have cyrix branded chip that's able to run at 133 reliably. I have been through a few but it was always just the IBM'S were long term stable at 120, and of those IBMs I could never get one to run at 133. I could often complete post, and on the m918 I was able to actually run a few dos benchmarks, however once the cpu warmed up it was all over. I assume with a peltier or ice I could do it. I don't think it was a motherboard issue at 66mhz as I was able to use fairly fast timings on the 918 at 60mhz, it seemed wholly a thermal issue.
Indeed! That's why I'm so surprised. I have a few other Cx5x86 chips and in the past I was able to OC one labeled 120MHz to 133MHz using the M921. Would it be Windows stable? I didn't test it, and my R9250 is not 66MHz capable anyway. It would be interesting to redo the tests but it takes so much time...
I am actually in the other boat right now, trying to find an AMD to run at 3x60 to get all I can out of this thing 😀
I think they're easier to find than the Cx5x86 4x 133MHz but it's just a matter of luck. I had an ADZ that was 180MHz capable but not anymore, so I only have one that still supports 180MHz@4v and it's an ADW A9630 BPG. Technically the ADZ labelled chips should withstand higher temperatures and thus be easier to OC AFAIK.
Was it tested in a case with the lid on? Your experience may be different with the lid on during the summer.
It's sitting on a test bench. I noticed this CPU is very sensitive to temperature so I'm using a beefy 462 heat sink with a slow 120mm fan on top of it to keep it cool. It's not even lukewarm.
It doesn't matter how beefy a heatsink or how fast the fan is - they won't reduce the CPU core temperature to below the ambient temperature. Putting the assembly into a case increases the ambient temperature, usually to around 32 C. Running a cased system in the summer further increases the ambient temperature immediately surrounding the CPU.
Any game which uses Direct3D is sufficient for testing purposes. Try Outlaws with the later patches.
Normally you run the Cyrix register settings prior to putting Win9x to prevent the system from hanging or rejecting the setting you are trying to apply.
feipoawrote on 2025-02-09, 22:25:bertrammatrix:
What BIOS are you running with that single 64 MB module? If you are using any of the BIOSes later than 5/6/96, ch […] Show full quote
bertrammatrix:
What BIOS are you running with that single 64 MB module? If you are using any of the BIOSes later than 5/6/96, check your DOS Quake score with the 64 MB chip, then test it again with a 32 MB chip (or even 96 MB total). If your Quake score is substantially lower with 64 MB total, you have run into the "64 MB bug".
I concur, Outlaws is definitely playable and enjoyable on a hyper 486.
Yes, some games don't like particular Cyrix enhancements, but you may also discover that reducing your CPU operating frequency may let some enhancements still work. Most unfortunate are the games which don't like FP_FAST. I think Turok was one such game.
Yeah the FP_FAST usually makes a noticeable difference if it is able to be on.
I've been through all the bioses available on both my boards (v1.5 and v3.4) since we've last discussed this. I found ALL bioses had the same bug on both boards Except that one. I tried to push for higher ram and / or cache speed settings every time and didn't seem to achieve anything no matter which bios I used, so I conclude that one version without the bug is the best, unless of course my hardware is glitchy and some boards may be able to run faster settings with some bios, which can certainly be the case
... unless of course my hardware is glitchy and some boards may be able to run faster settings with some bios, which can certainly be the case
May be. This is something I need to test for again. It may be an edge case situation. My board can do 180 Mhz with 1024K L2, and 128 MB EDO with 2-1-2 SRAM and DRAM at 1 ws read, 0 ws write. It may have been that I needed to set DRAM to 2 ws with the 5/6/1996 BIOS, but I don't recall.
Plan your life wisely, you'll be dead before you know it.
