I'd rather not risk the 133 chip though, would hate if it burned up @ 150Mhz. These chips are hard to find.
I do have a MediaGX board, but I never got it to post.
A Socket 3 to Socket 7 adapter would be sweet though! 😁

feipoa wrote:

I guess the closet thing to a stable Cyrix 5x86-150 would be a MediaGX chip. National got those up to 300 MHz. I wonder if it is possible to create some kind of MediaGX-to-socket3 interposer/converter board?

What about the Geode? That was based on the MediaGX before AMD bought it and made it Athlon based and was clocked even higher.

If looking into making an adapter/converter, I'd start with the original MediaGX, then try for GXm, or GXi. I'd leave the AMD and more advanced chips out of it. We'd want to put the chip into a mode which bypasses the on-chip graphics, sound, and the memory controller. I've used the GXm with off-chip PCI Graphics and sound without issue.

Are there any systems that use MediaGX chips which can bypass all integrated features? Is that even possible according to the datasheets?

Anonymous Coward wrote:

Are there any systems that use MediaGX chips which can bypass all integrated features? Is that even possible according to the datasheets?

I do not know to what extent they are bypassed, but the MediaGXm system I have setup allows you to disable the Cyrix IDE, Sound, and Video within the BIOS. I do not know to what extent they are really disabled architecturally or how much the Cyrix chipset is playing a role. The onboard Cyrix chipset, the Cx5530, might be largely an IDE controller.

If history is anything to go by, then getting a MediaGX working in a socket3 board is probably not going to work unless Cyrix intentionally designed it to do so. If you recall the intel 80186 (another highly integrated chip for its time), there were some changes made to the integrated components which made it impossible to design a 100% compatible PC around it. You would think if the MediaGX were capable of working on a 486 board that somebody would have released an upgrade chip based around it.

Your 486 is amazing!

feipoa wrote:

In reality, and for the really determined individual, you can probably OC a Cyrix 5x86 to 150 MHz, but you'd need to put in place a variable resistor on your motherboard's voltage regulation circuit to find the optimum voltage for which the CPU didn't overheat/crash. I'd buy the most expensive heatsink compound you could find for that extra 5% of cooling power, and a somewhat rare 3-clipped Socket 370 heatsink/fan. I say rare because I've only ever tested one 3-pronged clip style that will clip successfully onto a socket 3 ZIF socket. I have 2 now. Most will initially grab, but then slip off. An extra pivot point is needed in the clip. Unfortunately, most motherboards have capacitors and other SMD parts in the way, except for the MB8433-UUD.

Could you please post a photo of the clip style that does work ?

Just a point of clarification - when you say 486 you don't mean CPU,you mean class? Isn't a 5x86 an upgrade path from the 486, so technically a generation above the 486 in the same way that a K6 is a generation above the Pentium, using technically the same socket with a higher FSB? So if there was a super socket 3, that's what a 5x86 would use? I would have thought an overclocked DX4-120 was the fastest legitimate socket 3 CPU. If I'm showing naivety then I'm happy to find out what the deal is.

brassicGamer wrote:

Just a point of clarification - when you say 486 you don't mean CPU,you mean class? Isn't a 5x86 an upgrade path from the 486, so technically a generation above the 486 in the same way that a K6 is a generation above the Pentium, using technically the same socket with a higher FSB? So if there was a super socket 3, that's what a 5x86 would use? I would have thought an overclocked DX4-120 was the fastest legitimate socket 3 CPU. If I'm showing naivety then I'm happy to find out what the deal is.

The AMD 586 is just a 486 but the Cyrix 586 is a bit more advanced, most motherboards wont turn on most of the extra features though and some features like branch prediction only works ok in the later revisions of the Cyrix 586 as I understand it.

Branch prediction works with both revisions of the Cyrix 5x86 in DOS. Stepping 0, Revision 5 is pseudo-stable with branch prediction in Windows. The stability also seems frequency-dependent. The talk has beent hat Stepping 1, Revision 3 is stable in Windows, however I have yet to perform a detailed study for this case. It could also be that some work with branch prediction, and some do not. A lot of the special features are frequency-dependent, especially at 133 MHz.

When I say the '486', I mean the fastest non-Pentium chip, whatever it may be, which functions properly in a socket 2/3 motherboard. This definition varies largely by the individual user and it is the one I have adapted for myself. You all may decide on your own definition.

PARKE: Some Socket 370 CPU clips have 3 mounting holes per side of the clip. These can be used in 2-pronged (per-side) socket 3 motherboards. You can also use a Z-clip CPU clip. As an update to post, I was unable to get a Cyrix 5x86 running stable in Windows at 150 MHz. DOS was stable for awhile, but once the chip heats up, it starts to fail after some time. At 150 MHz, I had to use 4.0 V on a QFP IBM 5x86-100.

I haven't updated this link in some time. I decided it would be a good idea to upload the stable CMOS settings, POST images, and DOS/Windows Quake results for both of my Cyrix 5x86-133 systems.

Biostar MB-8433UUD v2.0
IBM 5x86-133/2x @ 3.70 V (QFP chip on interposer)
1024 KB write-back L2 cache (cache mod required)
64 MB FPM 60 ns
Win95c/WinNT4
UM8881F/UM8886BF
66 MHz FSB

Matrox G200 16 MB
Voodoo2 12 MB
Promise Ultra100 TX2
Maxtor DiamondMax Plus 9 Ultra133 HDD - 80 GB - 7200 RPM - 8 MB
Intel EtherExpress Pro 100 ISA
Audican32 Plus (Yamaha YMF718-S - OPL3-SAx) w/NEC XR385

UDMA3 IDE CD-ROM
3.5" CF Card adapter connected to Promise Ultra100 (unused)
Logitech PS/2 MouseMan M-S38

Cachechk
L1 = 274 MB/s
L2 = 101.6 MB/s
RAM Read = 69.6 MB/s (60 ns)
RAM Write = 92.6 MB/s (45 ns)

Speedsys
L1 = 198.4 MB/s
L2 = 73.5 MB/s
RAM = 56.5 MB/s

GLQuake at 800x600x16 shows 29.1 fps, but sometimes goes up to 29.3 fps.
-nosound -nocdaudio -nonet -nomouse -nojoy -wavonly

Stable CMOS settings for 64 MB. For 128 MB, you might need to add another read WS.

DTK-PKM-0033S
Cyrix 5x86-133/4x @ 3.70 V
1024 KB write-through L2 cache
128 MB FPM 60 ns
Win95c/WinNT4
SiS 496NV / SiS 497NU
33 MHz FSB

Voodoo3 3000 16 MB
Adaptec 2940U2W Ultra2 SCSI
Seagate Ultra320 SCSI HDD - 146 GB - 10K RPM - ST3146707LW
3Com 3c515-TX 10/100Base-TX
Creative AWE64Gold with 28 MB SIMMConn

SCSI2 CD-ROM - NEC CDR-1910
3.5" CF Card adapter connected to onboard IDE (unused)
Logitech PS/2 MouseMan M-S38 (PS/2 circuit & BIOS mod required)

Cachechk
L1 = 273 MB/s
L2 = 93.1 MB/s
RAM Read = 55.5 MB/s (75 ns)
RAM Write = 69.3 MB/s (60 ns)

Speedsys
L1 = 191.1 MB/s
L2 = 59.0 MB/s
RAM = 44.0 MB/s

GLQuake at 800x600x16 shows 28.0 fps.
-nosound -nocdaudio -nonet -nomouse -nojoy -wavonly

For some reason, write-back L2 cache was not stable with 128 MB, so I am using write-through L2 cache.

I suppose the OP is attempting to create a system with maximum performance, while retaining full functionality, but if I was going to maximize performance of a system dedicated to retro purposes, then I would probably remove anything that didn't need to be there, after the fact. For instance, after all software is installed, I would remove the floppy drive and disable it's controller, remove any CD-ROM drives and disable any secondary IDE controllers, remove any networking hardware, disable all COM ports unless you're using a serial-mouse, etc. and disable PS/2 unless you're using a PS/2 mouse.*

*I'm curious how using a PS/2 mouse, which is typically on IRQ 12 impacts the performance of a system as opposed to say, a serial mouse on COM1 or COM2 which have a much higher priority in the IRQ line.

I have never compared serial vs. PS/2 mouse with benchmarking. It would be interesting, however, I suspect the impact is minimal.

I recently swapped cases on my IBM 5x86-133/2x+Voodoo2 system. The trap door prevents my kids from hitting the power button.

A few shots of the thermally epoxied fan on QFP.

Your alu cooler is nice. It's from a T-Bird or Palomino/Thunderbird?

Did you put some thermal glue capable paste between chip and cooler?

Yes, I had to use thermal epoxy to secure the heatsink onto the CPU because the interposer PCB sticks too far and blocks the Z clip area. Another view of that PCB is here, Re: The World's Fastest 486
I do believe that heatsink came from some AMD system, but forget which. Those two photos are actually of two different heatsinks, that is, the one inside the case is not the one on the desk. I have 3 of these IBM 5x86c QFP chips which I thermally epoxied heatsinks to. I have a 4th which I left bare, and a 5th which died after my 150 MHz experiments.

Well the fastest possible 486 machine would be an industrial board with several 486 cards in it.

Granted it may not play videos games very well. But If we're talking terms of raw power that be as fast as you can get. It's entirely possible to have a 486 number crunching farm in one machine.