I think he means the 4x chips that don't have "4x" printed on them. They look exactly like non-4x chips, both for versions with and without OEM heatsink.

magicmanred wrote on 2024-07-30, 18:00:

Very interesting. Been trying to find a 4x variant. […]
Show full quote

feipoa wrote on 2024-07-30, 15:16:

You definitely need to run some Quake and Windows tests to unfold all the issues awaiting you with Cyrix 5x86 -100 and -120 chips marked at 4x when operating at 133 MHz. Your best chances for stable 133/4x operation are with the non-speed marked chips with 1996 date codes.

Very interesting.
Been trying to find a 4x variant.

What do you mean by non-speed marked?
Can you post an example image?

Do you mean something like this:
https://cdn.cpu-world.com/CPUs/5x86/S_C ... speed).jpg

Yes, aside from 133/4x marked chips, that CPU you've shown has returned the best luck with for 133/4x operation, as long as the datecode is from 1996.

feipoa wrote on 2024-07-30, 15:16:

You definitely need to run some Quake and Windows tests to unfold all the issues awaiting you with Cyrix 5x86 -100 and -120 chips marked at 4x when operating at 133 MHz. Your best chances for stable 133/4x operation are with the non-speed marked chips with 1996 date codes.

Just got around to it, my -100gp and two -120gp chips all seem to work great at 133. Tested with Quake and Doom. I had lsser, fp_fast, and btb_en all enabled, no stability issues. This is also at the stock 3.45v. Active cooling is probably a good idea though. My passively cooled 120's (with OEM Cyrix heatsink) get a bit warm to the touch. However, my 100 with aftermarket heatsink and fan stays cool as a cucumber!

magicmanred wrote on 2024-07-30, 18:00:

Do you mean something like this:
https://cdn.cpu-world.com/CPUs/5x86/S_C ... speed).jpg

I wouldn't bother trying to find one of those blank ones. I've never seen one myself, and I've collected a lot of Cyrix 5x86's over the years. I own three 4x chips, and know a guy who owns another two. All of them are marked -120gp or -100gp.

willmurray461 wrote on 2024-08-02, 22:15:

I wouldn't bother trying to find one of those blank ones. I've never seen one myself, and I've collected a lot of Cyrix 5x86's over the years. I own three 4x chips, and know a guy who owns another two. All of them are marked -120gp or -100gp.

Thanks for the tip!

Are you, or anyone you know selling one? Would love to have one to tinker with on my ASUS PVI-486SP3.
I use a PCI Voodoo1 Display card paired with a VLB 2MB Tseng Labs ET4000/W32P, as well as an ISA AWE32 so upping the FSB much over 40Mhz may start being a concern.
But 33x4 would definitely be something that would agree with everything in my system 😁

I am currently doing 160Mhz (40x4) with an AM5x86 133ADW fully stable with 1MB MB cache, so 40Mhz fsb works great.

If I'm being completely honest, I don't think a Cyrix 5x86-133 is going to bring that much value to your system, especially since you already have a working AMD X5-160.

Also, market price for Cyrix 5x86-133's seems to be like $700+ 😬. If you find one on your own it won't be that expensive, but if you're buying from a collector expect to pay a lot.

In my opinion, 5x86-133's don't really make much sense in socket 2/3 systems, especially if they have 40MHz bus capability. If your 486 system is locked to a 33MHz bus and is not Pentium Overdrive compatible, then the Cyrix 5x86-133 has a decent lead over all the alternatives. However, in 40MHz or POD compatible systems, the Cyrix 5x86-133's lead diminishes greatly.

If you want to try out Cyrix stuff, it will be much cheaper to try to find a 120MHz model. The performance is very close to the 133 (and maybe even better in some scenarios due to the faster bus). Additionally, if you find an early -120GP (~week 42 1995 or earlier), there is a good chance it will be S1R3, meaning you can run branch prediction stable in Windows, which is impossible for the -133GP since it only exists in the S0R5 variety.

willmurray461 wrote on 2024-08-09, 18:58:

If I'm being completely honest, I don't think a Cyrix 5x86-133 is going to bring that much value to your system, especially sinc […]
Show full quote

If I'm being completely honest, I don't think a Cyrix 5x86-133 is going to bring that much value to your system, especially since you already have a working AMD X5-160.

Also, market price for Cyrix 5x86-133's seems to be like $700+ 😬. If you find one on your own it won't be that expensive, but if you're buying from a collector expect to pay a lot.

In my opinion, 5x86-133's don't really make much sense in socket 2/3 systems, especially if they have 40MHz bus capability. If your 486 system is locked to a 33MHz bus and is not Pentium Overdrive compatible, then the Cyrix 5x86-133 has a decent lead over all the alternatives. However, in 40MHz or POD compatible systems, the Cyrix 5x86-133's lead diminishes greatly.

If you want to try out Cyrix stuff, it will be much cheaper to try to find a 120MHz model. The performance is very close to the 133 (and maybe even better in some scenarios due to the faster bus). Additionally, if you find an early -120GP (~week 42 1995 or earlier), there is a good chance it will be S1R3, meaning you can run branch prediction stable in Windows, which is impossible for the -133GP since it only exists in the S0R5 variety.

Very good points. I can definitely see what you're talking about.
I do have a Cyrix 5x86 100GP sitting on my shelf on display. Haven't popped it in to play with. I felt like it would be worth maybe selling it and getting a 120 to play with, but maybe I'll pop it in and see if it can do 40x3 and/or branch prediction.

1993 IBM PS/1

Specs:
486dx2-50mhz
32mb RAM
64kb cache
IBM IDE hard drive

I'm getting back again to tinkering with my old 486 motherboards and really love this benchmark!

Finally was able to get SpeedSys to run on the PS/1 2033 upgraded to a DX4-100 using an ODP:

Considerable upgrade vs the same CPU running at 75Mhz with the 25Mhz FSB:

Stock Pc Chips M912 with stock AMD 5x86 133 (4x33), no heatsink and modded 2x16Mb (EDO to FPM w parity). Nothing out of the ordinary.

I could manage to max out my Isa ide performance further on my 386sx board w/ TI486 CPU.

Badly camera timing but buff read can be identified.

magicmanred wrote on 2024-08-11, 04:37:

willmurray461 wrote on 2024-08-09, 18:58:

If I'm being completely honest, I don't think a Cyrix 5x86-133 is going to bring that much value to your system, especially sinc […]
Show full quote

If I'm being completely honest, I don't think a Cyrix 5x86-133 is going to bring that much value to your system, especially since you already have a working AMD X5-160.

Also, market price for Cyrix 5x86-133's seems to be like $700+ 😬. If you find one on your own it won't be that expensive, but if you're buying from a collector expect to pay a lot.

In my opinion, 5x86-133's don't really make much sense in socket 2/3 systems, especially if they have 40MHz bus capability. If your 486 system is locked to a 33MHz bus and is not Pentium Overdrive compatible, then the Cyrix 5x86-133 has a decent lead over all the alternatives. However, in 40MHz or POD compatible systems, the Cyrix 5x86-133's lead diminishes greatly.

If you want to try out Cyrix stuff, it will be much cheaper to try to find a 120MHz model. The performance is very close to the 133 (and maybe even better in some scenarios due to the faster bus). Additionally, if you find an early -120GP (~week 42 1995 or earlier), there is a good chance it will be S1R3, meaning you can run branch prediction stable in Windows, which is impossible for the -133GP since it only exists in the S0R5 variety.

Very good points. I can definitely see what you're talking about.
I do have a Cyrix 5x86 100GP sitting on my shelf on display. Haven't popped it in to play with. I felt like it would be worth maybe selling it and getting a 120 to play with, but maybe I'll pop it in and see if it can do 40x3 and/or branch prediction.

If you want more than 120Mhz with a Cyrix 5x86 but don't want to loose performance due to a 33Mhz bus, there is the option to mod your board for custom FSB by replacing the crystal as described by me here: Custom FSB (e.g. 46Mhz) on a 486 VLB mainboard Running the l2 cache with zero additional waitstates on a 45Mhz FSB is possible (on SIS471 chipset, don't know about others) but be prepared to swap your sram chips a lot until you find some that work at this speed 😁

But you have to keep in mind that this is not a 100% clean solution until you fix the problem with the timer being messed up...

A few benchmarks from my 486 with an IBM 5x86c running at 134 (3x44,52) Mhz:

Stesch wrote on 2025-06-16, 17:29:

But you have to keep in mind that this is not a 100% clean solution until you fix the problem with the timer being messed up...

I experimented with this 15 years ago and determined that I could not add more than about ~1 MHz FSB before the floppy drive stopped reading/writing correctly. Are you able to read/write an entire floppy at 44.5 MHz FSB?

More ideal is to replace the whole PLL with a custom job. I've been meaning to bodge wire up something on an AL486V-D, but it is low on my list. Probably start with something like this: 386 Clock Generator Replacement or Re: Project: Full Can Clock Oscillator Replacement

feipoa wrote on 2025-06-16, 23:20:

Stesch wrote on 2025-06-16, 17:29:

But you have to keep in mind that this is not a 100% clean solution until you fix the problem with the timer being messed up...

I experimented with this 15 years ago and determined that I could not add more than about ~1 MHz FSB before the floppy drive stopped reading/writing correctly. Are you able to read/write an entire floppy at 44.5 MHz FSB?

More ideal is to replace the whole PLL with a custom job. I've been meaning to bodge wire up something on an AL486V-D, but it is low on my list. Probably start with something like this: 386 Clock Generator Replacement or Re: Project: Full Can Clock Oscillator Replacement

Hi feipoa,

I only tested if the dir command works with one of my floppys and it worked just fine, I will do some more tests today after work and report back. Im not sure if this is relevant, but I use an slower crystal (12.75MHz) instead of the stock 14.31818Mhz one in combination with the 50MHz FSB jumper setting to achive 44.5 MHz FSB, perhaps that makes a difference?

Replacing the PPL sounds very interesting, if you get around to it on your AL486V-D I would be happy to see a thread about it 😀

I dug up my old notes.

For most common PLL's found on socket 3 boards,

FSB / multiplier
20 / 1.3968
25 / 1.7460
33 / 2.3280
40 / 2.7937
50 / 3.4921
60 / 4.1905
66 / 4.6561

With the Cyrix 5x86, Itested:

19.2000 MHz * 2.3281 = 44.7 MHz FSB, or 134 MHz to the Cx5x86
19.6608 MHz * 2.3281 = 45.8 MHz FSB, or 137 MHz
20.0000 MHz * 2.3281 = 46.5 MHz FSB, or 140 MHz

With the Am5x86, I tested:

15.36 MHz * 2.3281 = 41.9 MHz FSB, or 167 MHz to the Am5x86 [I marked this test with an asterik, implying it worked]
20.00 MHz * 2.79365 = 55.9 MHz FSB, or 167.6 MHz
15.36 MHz * 2.79365 = 43.0 MHz FSB, or 171.6 MHz [I noted that this test was "OK"]

With the POD, I tested:
15.360 MHz * 2.3281 = 36 MHz FSB, or 89.4 MHz to the POD [I marked this test with an asterik, implying it worked]
16.384 MHz * 2.3281 = 38 MHz FSB, or 95.4 MHz
16.000 MHz * 2.3281 = 37 MHz FSB, or 93.1 MHz

I have written in my notes, "cannot exceed 15.36 MHz crystal". From the tests shown above, even 16.00 MHz didn't have all the functions working correctly. This means that only approx. 1.0418 MHz above (or below?) the standard 14.31818 MHz can be used. But there is a range I did not test for, namely a delta between 1.0418 MHz and 1.65 MHz. Your delta value of 1.57 MHz falls within this range, so, if it works, it may be on the edge. I don't have any crystals lower than 14.31818 MHz, so I am curious if 12.75 MHz is really a working solution. T

From what I recall, "dir" to display floppy contents still worked. Try copying a 800 KB file from the floppy disk to the HDD and try writting a different ~800 KB file to the floppy drive. Also, you may want to try running an executable file from the floppy disk.

If you have your keyboard controller operating at some divisor of the PCI bus, e.g. PCI/3, check for hang-ups with the keyboard. Or did you set your keyboard to a fixed 7.16 MHz?

If the system timer is messed up, I think your benchmark results will also be incorrect. It is this, plus the floppy or keyboard causing issues which turned me away from this approach. More ideal is to find a socket 3 which contains a DIP-14 for a crystal oscillator, then use one of the previously linked solutions. Alternately, running an IBM 5x86c at 2x66 on an MB-8433UUD is another solution.

feipoa wrote on 2025-06-17, 08:24:

From what I recall, "dir" to display floppy contents still worked. Try copying a 800 KB file from the floppy disk to the HDD and try writting a different ~800 KB file to the floppy drive. Also, you may want to try running an executable file from the floppy disk.

If you have your keyboard controller operating at some divisor of the PCI bus, e.g. PCI/3, check for hang-ups with the keyboard. Or did you set your keyboard to a fixed 7.16 MHz?

I will try that when I am at home and report back. Its an VLB only board (ASUS VL/I-486SV2GX4), so no PCI divisor for me. Also, there is no BIOS option to change the keyboard controller speed. The keyboard works fine tho, I've played around 8-9 hours now on this system (mostly C&C Red Alert and fatal racing) and didn't experience any hang-ups so far. When I tried the undocumented 66MHz FSB setting some time ago for 133Mhz (like you did for your IBM 5x86c ), the keyboard controller stopped working indeed 😁

Sometimes there's a fast/slow setting where fast means small divider, slow means larger divider. Also turning off Typematic rate setting usually helps. Sometimes you have a choice between internal chipset KBC and external KBC.... which won't do anything useful for ext if the pads are empty. Some boards are just marginal with certain keyboards, so a fail with one KB doesn't necessarily mean a fail with all keyboards.

feipoa wrote on 2025-06-17, 08:24:

Try copying a 800 KB file from the floppy disk to the HDD and try writting a different ~800 KB file to the floppy drive. Also, you may want to try running an executable file from the floppy disk.

I am back from work and have just tested the floppy drive by doing the following:

- format a floppy
- copy lotus III (small .cfg and .exe, around 1.2 Mb big .dat file) to the floppy
- run lotus directly from the floppy
- copy the content of the floppy disk back to the HDD into a new test folder and run it there

All tests worked without any problems, aside from the long loading / transfer times (but I think this is due to the 'slow refresh' option in BIOS). I also tried running Planet X3 directly from the original floppy and it works too. It seems the floppy drive is comfortable with its environment 😀

Lotus III:

Planet X3 VGA:

Edit: I've disabled the slow refresh option in the BIOS and loading times are now much shorter. So it was this option indeed...

Do you have crystals of lower values? It would be interesting to see at which delta the floppy drive, or other functions, cease working correctly. On my MB-8433UUD, going from 14.31818 MHz to 16.00 MHz messed up the floppy drive.

I think it would also be helpful to determine how much your benchmark results are off by when using the 12.75 Mhz crystal. You can get some idea by comparing various results with the 14.31818 MHz crystal to that of the 12.75 Mhz crystal.

Does your BIOS track the day time correctly?