MSxyz wrote on 2025-07-14, 19:11:

Stesch wrote on 2025-07-14, 07:16:

From my experience, BTB is not 100% stable on S0R5 CPUs, even under DOS. Most games will run fine for several hours without any problems, but others will crash/hang much sooner, e.g. Transport Tycoon Deluxe. S1R3 CPUs worked fine with BTB enabled for me, but they tend to overclock less than their S0R5 counterparts. Do you enable only these three registers or more when turning BTB on?

Only those three. As for overclocking, I usually run this CPU at 120MHz, stock voltage, without issues.

Yes, the IBM ones are quite good overclockers (but the Cyrix also aren't as bad as some people tell, at least in my experience). Never have seen an IBM 5x86c with Stepping 1 Revision 3 before, I think it is quite rare, so take good care of it 😀 I think you could also enable DTE, memory read bypassing and BWRT on your IBM without issues (and if your chipset supports it, also LINBRST)

bertrammatrix wrote on 2025-07-12, 04:12:

MSxyz wrote on 2025-06-30, 16:11:

Recently, there was on eBay a 5x86 120 MHz... the seller claimed it supported the 4X multiplier and asked 550$ as a starting price for it (it was put on auction). I forgot to bookmark the page, but since I can't find it anymore, I guess -in the end- it was sold for that exorbitant price.

I've a couple of 5x86-100 that run fine at 120 MHz and also boot to DOS at 66x2, although the boards I have don't seem to be very stable at that frequency. One is step 0 rev 5 and it has a green metal heatsink with the Cyrix logo, the other is step 1 rev 3. and it has a blue metal heatsink with the IBM logo. Never had the pleasure to find a 4X variant.

Out of curiosity - set them to 3 x 50 instead at 4 volts and see if they post. Looks like a few QFP ones can do this, no word of any success with PGA so far

Back in the day on my BEK/UMC/VLB board, my 100GP PGA could be set to 3x50 and POSTed, but I never back then figured out what it was actually doing. Ran at 2x50 fine, ran at 3x40 fine, ran at 2x60 pretty good with the odd glitch. POST reported 125Mhz when set to 3x50 and benchmarks seemed to land between 3x40 and 2x60. I didn't have/find the better at getting CPU clock utils at the time. The apparent conclusion from results obtained was that it was running at 2.5x ???? Which seems super weird. However there is a possibility that it was signalling so bad it was "fuzzing" the caches, such that they needed to flush and refill frequently enough to bring performance way down from what should be expected for 150Mhz. I have not got around to setting up an extreme performance 486 board again more recently to see if prior experience can be replicated... I do not recall the voltage employed, it could have been 3.8, but I think 3.6 was where I usually was.

BitWrangler wrote on 2025-07-15, 12:32:

bertrammatrix wrote on 2025-07-12, 04:12:

MSxyz wrote on 2025-06-30, 16:11:

Recently, there was on eBay a 5x86 120 MHz... the seller claimed it supported the 4X multiplier and asked 550$ as a starting price for it (it was put on auction). I forgot to bookmark the page, but since I can't find it anymore, I guess -in the end- it was sold for that exorbitant price.

I've a couple of 5x86-100 that run fine at 120 MHz and also boot to DOS at 66x2, although the boards I have don't seem to be very stable at that frequency. One is step 0 rev 5 and it has a green metal heatsink with the Cyrix logo, the other is step 1 rev 3. and it has a blue metal heatsink with the IBM logo. Never had the pleasure to find a 4X variant.

Out of curiosity - set them to 3 x 50 instead at 4 volts and see if they post. Looks like a few QFP ones can do this, no word of any success with PGA so far

Back in the day on my BEK/UMC/VLB board, my 100GP PGA could be set to 3x50 and POSTed, but I never back then figured out what it was actually doing. Ran at 2x50 fine, ran at 3x40 fine, ran at 2x60 pretty good with the odd glitch. POST reported 125Mhz when set to 3x50 and benchmarks seemed to land between 3x40 and 2x60. I didn't have/find the better at getting CPU clock utils at the time. The apparent conclusion from results obtained was that it was running at 2.5x ???? Which seems super weird. However there is a possibility that it was signalling so bad it was "fuzzing" the caches, such that they needed to flush and refill frequently enough to bring performance way down from what should be expected for 150Mhz. I have not got around to setting up an extreme performance 486 board again more recently to see if prior experience can be replicated... I do not recall the voltage employed, it could have been 3.8, but I think 3.6 was where I usually was.

Are you sure it was a Cyrix 5x86 and not one of the IBM branded ones? The Cyrix 5x86GP PGA CPUs rated at 100MHz can often reach 120MHz, perhaps if you're extremely lucky 133 (2x66MHz or 4x33MHz for those with the rare x4 multiplier option) but 150MHz would surprise me... Even most IBM 5x86c can't reach 150MHz without major overvolting (considering that they are rated for 3.3 volts and not 3.6/3.7 like the Cyrix ones).

Definitely a Cyrix. Was an early one I think due to having more lsser and btb problems than most, but that might have been the motherboards beta support for it.

Stesch wrote on 2025-07-15, 06:04:

MSxyz wrote on 2025-07-14, 19:11:

Stesch wrote on 2025-07-14, 07:16:

From my experience, BTB is not 100% stable on S0R5 CPUs, even under DOS. Most games will run fine for several hours without any problems, but others will crash/hang much sooner, e.g. Transport Tycoon Deluxe. S1R3 CPUs worked fine with BTB enabled for me, but they tend to overclock less than their S0R5 counterparts. Do you enable only these three registers or more when turning BTB on?

Only those three. As for overclocking, I usually run this CPU at 120MHz, stock voltage, without issues.

Yes, the IBM ones are quite good overclockers (but the Cyrix also aren't as bad as some people tell, at least in my experience). Never have seen an IBM 5x86c with Stepping 1 Revision 3 before, I think it is quite rare, so take good care of it 😀 I think you could also enable DTE, memory read bypassing and BWRT on your IBM without issues (and if your chipset supports it, also LINBRST)

The other two 5x86s I've are from Cyrix (one with the green heatsink attached, one without) and they're also good at 120MHz. Step 0 Rev 5 both of them, manufactured near the end of 1995.

My favourite motherboard with these CPUs is the M919 rev 3.4. I found a new old stock a couple of years ago with the original cache module. The other two mobo I use with these chips are an Asus 486SP3 and a Zida 4DPS . The Asus one has the clock generator capable of 66 MHz operation and I've tried running it with an Intel P24C with some success. None of my 5x86s, however, are stable at 66x2... They barely reach the DOS prompt, on a cold boot, and then crash.

The M919 is definitively the fastest of the bunch, especially with a VLB video card and enabling linear burst in the BIOS. If I had to rank them according to speed with popular DOS games like DOOM or Quake it would be:

M919 VLB > 486SP3 VLB > 486SP3 PCI > 4DPS PCI > M919 PCI This holds true both at 100 and 120 Mhz. (The M919 also sets the PCI clock to 26.6 MHz when the CPU bus is at 40 MHz, but even using the 'hot swapping trick' doesn't make it noticeably faster than the other two motherboards)

BitWrangler wrote on 2025-07-15, 16:39:

Definitely a Cyrix. Was an early one I think due to having more lsser and btb problems than most, but that might have been the motherboards beta support for it.

Thats impressive! Must have been one of those "magical" CPUs 😁 My IBM 5x86c is currenty running @133Mhz (3x44.52MHz) with 3.58 volts just fine, but 150MHz seems unreachable. So far I've had no luck finding an SQFP variant... Not to mention the unicorn 133GP variant...

Stesch wrote on Yesterday, 06:18:

BitWrangler wrote on 2025-07-15, 16:39:

Definitely a Cyrix. Was an early one I think due to having more lsser and btb problems than most, but that might have been the motherboards beta support for it.

Thats impressive! Must have been one of those "magical" CPUs 😁 My IBM 5x86c is currenty running @133Mhz (3x44.52MHz) with 3.58 volts just fine, but 150MHz seems unreachable. So far I've had no luck finding an SQFP variant... Not to mention the unicorn 133GP variant...

Right now, on eBay, there's one person selling 4 brand new 5x86 120 in QFP package and at reasonable price, too. They're rather useless, unless someone comes up with a custom pga to qfp adapter (or recycle of one of those made for Amd 5x86... The pinout should be similar, unlike earlier Cyrix CPUs). Also, there's no guarantee that they will work at a higher frequency (or that they work at all...) and you're not going to discover it before soldering all them 208 pins to the adapter pcb. 😀

(PS: From the pictures the seller provided, the production code appears to be G5A8544B, and they seem to originate from Korea - So they should be step 5 CPUs manufactured in week 44/1995)

MSxyz wrote on Yesterday, 07:26:

Stesch wrote on Yesterday, 06:18:

BitWrangler wrote on 2025-07-15, 16:39:

Definitely a Cyrix. Was an early one I think due to having more lsser and btb problems than most, but that might have been the motherboards beta support for it.

Thats impressive! Must have been one of those "magical" CPUs 😁 My IBM 5x86c is currenty running @133Mhz (3x44.52MHz) with 3.58 volts just fine, but 150MHz seems unreachable. So far I've had no luck finding an SQFP variant... Not to mention the unicorn 133GP variant...

Right now, on eBay, there's one person selling 4 brand new 5x86 120 in QFP package and at reasonable price, too. They're rather useless, unless someone comes up with a custom pga to qfp adapter (or recycle of one of those made for Amd 5x86... The pinout should be similar, unlike earlier Cyrix CPUs). Also, there's no guarantee that they will work at a higher frequency (or that they work at all...) and you're not going to discover it before soldering all them 208 pins to the adapter pcb. 😀

(PS: From the pictures the seller provided, the production code appears to be G5A8544B, and they seem to originate from Korea - So they should be step 5 CPUs manufactured in week 44/1995)

Thanks for the info! Could you perhaps send me a pm with the link? I would love to get one or two of these, but I can't find them on ebay, even when selecting 'world wide'. Perhaps I need to improve my ebay search skills 😀

Nevermind, I've found it on ebay.com. No shipping to Germany tho 🙁

willmurray461 wrote on 2024-07-25, 13:35:

Does anyone have an IBM 5x86C made around early '96 that they can test to see if it has 4x?

I have several between 51 1995 and 12 1996. They are all 2x/3x only.

MSxyz wrote on Yesterday, 07:26:

From the pictures the seller provided, the production code appears to be G5A8544B, and they seem to originate from Korea - So they should be step 5 CPUs manufactured in week 44/1995)

On the contrary, they are Step 1, Rev 3. I think 44-1995 was the last for Step 1, Rev 3 Cyrix 5x86 CPUs. I have a few waiting transplant onto PQFP-PGA interposers. Unfortunately, that seller doesn't package the CPU's in trays, so I have a lot of unbending of pins to take are of.

feipoa wrote on Yesterday, 11:29:

On the contrary, they are Step 1, Rev 3. I think 44-1995 was the last for Step 1, Rev 3 Cyrix 5x86 CPUs. I have a few waiting transplant onto PQFP-PGA interposers. Unfortunately, that seller doesn't package the CPU's in trays, so I have a lot of unbending of pins to take are of.

My 5x86 with the green heatsink was made in week 45 and it's a step 5, so it appears that the next week there was at least one fab using the newer masks.

So, I take you've already purchased a few of them from that seller, right? Did you have the chance to try one of them already?

EDIT: I looked at my third 5x86, the one without heatsink, and it was made in week 41; I don't have the means to run it now, since most of my stuff is in storage due to moving to a new location, but I'm quite sure it was also a rev 5 chip. I could be wrong though... I've used it only once or twice not to damage the silkscreen installing a heatsink on top of it. I keep it as a showpiece. If anyone is interested, code on the back of the chip is G5L8541F.

I bought one CPU, tested it briefly, then bought a few more. They are S1R3.

Week 45 of 1995 started the S0R5. I have tested nearly a hundred chips and everything from Week 44 and under was S1R3 and everything from Week 45 and above was S0R5. On the other hand, I haven't tested one with markings of G5L8541F. Closest was G5L8541C and it was S1R3.

^^^ Here is what I got in the mail this week from that guy

Stesch wrote on 2025-07-14, 07:16:

From my experience, BTB is not 100% stable on S0R5 CPUs, even under DOS. …..R3 CPUs worked fine with BTB enabled for me, but they tend to overclock less than their S0R5 counterparts.

It’s odd that they reverted back to the S0 masks inless there were some significant yeild issues.

douglar wrote on Yesterday, 23:13:

The attachment IMG_2847.jpeg is no longer available

^^^ Here is what I got in the mail this week from that guy

49-1995. Interesting. I thought his all were 44-1995. Does the bottom side also indicate it being from Korea? Was yours shipped without a tray and with the CPU taped between two foam pads?

Will you be able to run the CPU to see if it is S0R5?

feipoa wrote on Today, 00:13:

douglar wrote on Yesterday, 23:13:

The attachment IMG_2847.jpeg is no longer available

^^^ Here is what I got in the mail this week from that guy

49-1995. Interesting. I thought his all were 44-1995. Does the bottom side also indicate it being from Korea? Was yours shipped without a tray and with the CPU taped between two foam pads?

Will you be able to run the CPU to see if it is S0R5?

Yes, Korea
Yes, in foam, pins slightly bent
He must have pets because my dog wanted to eat it in a bigly way.
I'm still in the market for a good QPF interposer, but I do want to try it out. Could I re-use one from an intel 486sx?

douglar wrote on Today, 01:31:

Yes, Korea Yes, in foam, pins slightly bent He must have pets because my dog wanted to eat it in a bigly way. I'm still in the m […]
Show full quote

Yes, Korea
Yes, in foam, pins slightly bent
He must have pets because my dog wanted to eat it in a bigly way.
I'm still in the market for a good QPF interposer, but I do want to try it out. Could I re-use one from an intel 486sx?

Ya, normally just a few pins are off, but I got one with nearly all the pins goofed up.

I once looked into using a 486SX interposer but realised that they are a few pins off. Even interposers with a Cyrix DX2 have way too many trace differences to workout well. Best to look out for an Am5x86 interposer. I'll see if my source has any left.

SX are way more than a few pins off, just to mention in case anyone thinks they are ambitious enough.

I have an AMD 5x86 I could sacrifice, but I’d rather not if I could avoid it.

BitWrangler wrote on Today, 02:16:

SX are way more than a few pins off, just to mention in case anyone thinks they are ambitious enough.

Sorry, my fingers don't always type what my brain is thinking. I meant to write, "the i486sx PQFP pin count is a few pins off from the cx5x86 PQFP pin count", namely PQFP-196 vs. PQFP-208 has a pin count difference of 12. Obviously, the pin mapping is way more than just a few trace differences.

My interposers all had i486DX4-100 WT chips on them. They work fine with Cx5x86 and Am5x86 CPUs. If you can find some from China, they are usually very cheap. I sent my contact a message of inquiry about more.