All documents have now been corrected. The results are not nearly as impressive anymore, but are perhaps more realistic.

I believe I have solved the IBM/Cyrix 5x86-133 voltage vs. stability issue, down to the last 10 mV.

Since W2K and Win98SE kept crashing after about 10-20 min. of playing a 256 kbps mp3 in Winamp, I decided to up the voltage from 3.70 V to 3.80 V. At 3.80 V, W2K would still eventually crash (BSOD), so I upped it to 3.90 V. At 3.90 V, I was able to play the mp3 for 1 hour without crashing. I did this twice. Success!

Now, in Win98SE, at 3.90 V, what I would do is play an mp3 for approx. 40 minutes (it didn't crash), but when I went to run Ziff-Davis Winbench9 - FPU WinMark, the benchmark would not complete and spat out some stop error. This error occurrence is very similar to what happened when the CPU was running at 4.0 V for more than about 20 minutes and I tried to run the benchmark (it got too hot). So as a compromise, I dropped the CPU voltage down to 3.85 V and have re-run the same series of long tests several times, and no crashes, and the benchmark program runs fine.

I wasn't even going to try a sub 100 mV step, but tried it as a last measure and it worked. So, 3.80 V was too little (frequency yield) and 3.90 was too high (thermal). 3.85 V was the sweet spot. I'd say, 3.84-3.86 V would probably be OK, but this really depends on how many BTUs your CPU fan is moving.

If you use a larger CPU fan, you might be able to get away with 3.90 V. Mine is pretty weak. Or if you put on one of those high power fans, like a 4 cm x 2 cm (as opposed to a 4 cm x 1 cm), you could probably get away with 4.0 V operation, which most motherboard's natively support.

Now, back into W2K to ensure that 3.85 V doesn't crash after 10-20 min. It has been 45 min. so far and no crash. The mystery Cyrix 5x86-133 is now something that is achievable to most retro overclockers.

Aside from the minimal production, I'm guessing that the yield of the marked Cyrix 5x86-133 chips was so low because they couldn't get enough of them to operate at 3.6V and didn't want to have a whole range of voltages printed onto their chips. What chip manufacturer would? They also had a big push for passive cooling and bragged about how little power their chips consumed.

Try that on your PCChips M919

My tabletop retro gear has been put away. Perhaps you can try it?

I already tried that and the IBM 5x86C-100 hangs at memory count

What voltage are you using?

I have two IBM 5x86C-100HF's and a Cyrix 5x86-120 that I tested this on and all of them worked fine at 133 MHz.

I wrote some graphical instructions on modifying a motherboard's voltage regulator circuit. See attachment.

Wow, thats a great guide.

Now it even not POST's, so the IBM 5x86C@133 is only for your Biostar mobo

@ Sebaz_ri

What voltage are you running the CPU at? The M919 natively supports 3.3V, 4V, and 5V.

• Set CPU to 4 V
• Add fan to CPU
• Set FSB to 66 MHz and multiplier to 2x
• In BIOS, set L2 cache to 3-2-2
• In BIOS, set DRAM Memory wait states to 1ws/0ws or 1ws/1ws
• Try a few PCI graphics cards, do not use VLB, Matrox G200 works well
• Have no other expansion cards (ISA/PCI) in the system

The M919 adds an auto 1:2/3 FSB-to-PCI divider, at least it does when the FSB is at 40 MHz. I'd assume it would do the same for 66 MHz, but this is untested. This means your graphics card is running at 44 MHz. From other testing, I know the G200 will run at even 50 MHz.

My MB8433-UUD board lets you manually select 1/1, 2/3, or 1/2 in the BIOS, but I do not know if the M919 has the ability for a 1/2 setting, but it should still work with a 2/3 setting. My MB8433-UUD works at 1:2/3 with 2 x 66 MHz.

Have you tried all the above?

• Set CPU to 4 V:
Already done
• Add fan to CPU
Already done
• Set FSB to 66 MHz and multiplier to 2x
Already done
• In BIOS, set L2 cache to 3-2-2
I don't have any L2 Cache
• In BIOS, set DRAM Memory wait states to 1ws/0ws or 1ws/1ws
I tried 3ws/3ws
• Try a few PCI graphics cards, do not use VLB, Matrox G200 works well
Tried with S3 ViRGE 325 and Alliance AT24
• Have no other expansion cards (ISA/PCI) in the system
I will need to try this

Additional Note: When i run the cpu in 5V mode it POST's but then it hangs at the memory count

The M919 isn't the most reliable motherboard. Do you have a Shuttle HOT-433 to test? Another Vogons memeber got a Cyrix 5x86-150 working on a HOT-433.

feipoa wrote:

The M919 isn't the most reliable motherboard. Do you have a Shuttle HOT-433 to test? Another Vogons memeber got a Cyrix 5x86-150 working on a HOT-433.

The M919 is the only 486-class mobo that i have!

feipoa wrote:

The M919 isn't the most reliable motherboard. Do you have a Shuttle HOT-433 to test? Another Vogons memeber got a Cyrix 5x86-150 working on a HOT-433.

The last HOT-433 I saw for sale, the seller wanted $200 for it. Good luck finding any 486 PCI motherboard cheap let alone one with the coveted UMC chipset.

Was it a HOT-433 revision 4 board on eBay about 2 months ago? It was going for $199. The seller was unrelenting, even with generous two-digit offers. I find the HOT-433's a bit awkward in terms of stability anyway. Maybe I keep getting lemons. 6 out of 6? naw. That board is kind of a write-off in my mind, but I was curious if they had worked out any bugs with that rev 4 board, but I wasn't crazy enough to pay $199 plus shipping (I think he wanted $49 to get it to Canada).

Anyone on here have a HOT-433 rev4 board?

feipoa wrote:

Was it a HOT-433 revision 4 board on eBay about 2 months ago? It was going for $199. The seller was unrelenting, even with generous two-digit offers. I find the HOT-433's a bit awkward in terms of stability anyway. Maybe I keep getting lemons. 6 out of 6? naw. That board is kind of a write-off in my mind, but I was curious if they had worked out any bugs with that rev 4 board, but I wasn't crazy enough to pay $199 plus shipping (I think he wanted $49 to get it to Canada).

Anyone on here have a HOT-433 rev4 board?

Yes, that's the one. As far as I'm concerned, even if you have to settle for a M919 for your fast 486/5x86 testing, and can get it cheap, then you should. Even though it's a low quality board with issues, it may be a long while before you find another 486/PCI/UMC board from any manufacturer for a reasonable price.

Anyone on here have a HOT-433 rev4 board?

I do (as you already know from my other posts). But what issues in particular should I be checking on?

I'm mainly interested to know if the PS/2 mouse port is functional and if there are any SCSI bus mastering issues. It would also be nice to know if it works with 512 KB cache and is stable in Windows 98SE/2000 (using PCI SCSI).

Here's a little update, not all IBM 5x86c-100HF CPUs upclock well to 133 MHz. Of the two I have, only one works very stably at 133 MHz and 3.85V, the other one is not stable at any voltage with 133 MHz. For the later, it is stable up to 120 MHz only.

I tested one of my Cyrix-branded Cyrix 5x86-120GP CPUs at 133 MHz and it is stable, but not at 3.85 V; it needed 3.90 V.

While it was prevously noted that a Biostar MB8433-UUD v3.0 works well with the 2x66 MHz FSB upclock, I have also recently tested a Biostar MB8433-UUD v2.0 and it also works well with a 66 MHz FSB. This is good news since the v2.0's are more widely available. Remember, 60ns FPM RAM seems to work much better than 60ns EDO RAM with 66 MHz operation.

Caviat: For some Biostar/CPU combinations at 133 MHz, the Cyrix feature RSTK_EN needs to be set off or you won't be able to boot into Windows. This took the better part of a day to troubleshoot. RSTK_EN is off by BIOS default. DTE may also need to be set off with some Biostar/CPU combinations to achieve long-term (several day) stability in Windows. This was only witnessed with the Cyrix 5x86-133/4X at 3.7 V. DTE is on by BIOS default.

I'm in the process of replacing all my soldered-in real-time clocks with DIP sockets so that the RTC's can be easily swapped out when it dies. RTC's don't usually last longer than 15 years, so their time is about up. I prefer the DIP sockets to the RTC hack job with the coin cell battery. I found out something interesting in this process. According to Speedsys,

Dallas DS12887A has the Y2K bug
Odin OEC12C887A has the Y2K bug
Dallas DS12887 does not have the Y2K bug

I have a Dallas DS12887+ I ordered from Digikey, but don't want to power it up to test it for Y2K bug or not. Maybe someone else with a DS12887+ can let us know if it has the Y2K bug.

After many days worth of testing, I have some additional conclusions to make about running an IBM 5x86c-100HF at 2x66 MHz. I have tried every possible voltage, fan cooling scheme, and enabling/disabling of all Cyrix and BIOS features.

1) For certain CPU/motherboard combinations, RSTK_EN must be disabled (=0).

2) This CPU is very sensitive to heat when being overclocked. This heat sensitivity is even feature-dependent (a tester's worst nightmare)! I could only enable LSSER (=0) if the case is off and there is adaquate cooling. Using a temperature element on the bottom side of the cermaic housing of the CPU, I estimate that LSSER can be enabled (=0) only when this temperature is less than or equal to 85 F. The system will crash if the temperature is any greater. This is ok if you want to run the CPU with the case off, but with the case on, I cannot get the CPU base temperature to less than 90 F after system warm-up. [Note that LSSER when set =0 is the enabled state, as opposed to all the other features where =1 is the enabled state.]

3) Ok, so how about running the CPU with LSSER off (=1)? This is doable, but after lengthy tests, it was discovered that LOOP_EN, BWRT, and DTE also need to be disabled (all set =0). These three other features don't do a whole lot for performance, so there is no major loss here. The good news is that BYP_MEM, LINBRST, and FP_FAST can still be enabled, however the real loss is with LSSER being disabled. LSSER has the largest impact on performance (about 7.5% for ALU and 11.5% for FPU), while FP_FAST helps only FPU, but by a big 18%

I tried the full swing of 3.55 - 3.95 volts to get LSSER working with less cooling, but it wasn't possible. If you want to run LSSER, leave the case off and put in a case fan, or use some alternate type of cooling. I have tried 2 case fans and a fat CPU cooler, but the case still needed to be off to achieve long-term stability.

4) The low-voltage crashing point is 3.54 V. I established this by running the CPU at 3.75 V then slowly turning down the voltage until I got a BSOD. I would say that the minimum running voltage is ~3.65 V. Above about 3.95 V, the CPU voltage becomes less stable on the multi-meter. This is either due to the motherboard's voltage regulator not being able to output more voltage (due to the other resistor's setting, the input bias voltage, or voltage drop in the regulator) or the CPU itself dropping more voltage (consuming more power and heating up).

5) Considering that a branded Cyrix 5x86-133/4x works with all these features enabled (except DTE), the situation may be related to the Cyrix's PLL circuit. It recently occured to me that the onboard PLL on the Cyrix chip may not be capable of reliably multiplying 66 MHz by 2x. Many PLL chips have a specified range for which they can multiply. For example, some common PLL chips used in motherboards can only reliably multiply an input single between 2 - 50 MHz. It is unknown what the Cyrix 5x86 PLL input requirements are, however it is unlikely the designers were considering a FSB above 50 MHz.

6) So, with LSSER disabled, you loose 7.9% from the overall system performance. Running a 66 MHz bus compared to a 33 MHz bus gains about 6% to the overall performance. It is, therefore, predicted that running with said features disabled that the system as a whole is very similar to a Cyrix 5x86-133/4x at 33 MHz. Now if you can run the system with better cooling, you can gain that 7.9% back. Please refer to the Ultimate 486 Benchmark Comparison and the Cyrix 5x86 Register Enhancements (in progress) for more information on this analysis.

I am currently running this CPU at 3.70 V and will run it for several months to establish long-term stability. I wonder if there is any way via CPU pin manipulation to trick a 3X Cyrix CPU into working at 4X? I think these are questions for Cyrix. Any ex-Cyrix employees lurking here?

EDIT1: I seem to be having some initial success with a branded Cyrix 5x86-120 at 133 MHz with LSSER enabled (stable for 2 hrs with full ALU/FPU load). The case is on and the voltage is set at 3.88 V. Does anyone know if the IBM 5x86C's used a smaller fab process compared to the Cyrix branded units? When I say fab process, I mean the length of the channel between the transistor's source and drain, in microns.

EDIT2: The CPU seems runs fine at 2x60 MHz (120 MHz) and 3x40 MHz, so any issues seem to be purely related to the high frequency and heat generation.

EDIT3:At 66 MHz FSB, my oscilliscope also measures 66 MHz. The waveform is clean sinewave.
Vpp = 0.84 V
Vrms = 1.50 V
Vmax = 1.92 V
Vmin = 1.05 V

At 33 MHz, the scope also measures 33 MHz. The waveform is a clean sinewave.
Vpp = 1.82 V
Vrms = 1.72 V
Vmax = 2.54 V
Vmin = 0.75 V

A lot of questionable stability from 2x66 MHz operation may not be solely due to the frequency of the signal to the CPU, but also related to the decreased voltage amplitude of the clock signal, not to mention the increase in Vmin. I'm not sure what the comparitor threshold is for socket 3 CPUs and what the rising/falling voltage of the clock signal need be. It may be that an op-amp pre-amplifier (voltage amplifier) needs to be put in place before running at 66 MHz FSB.

It may also be that manufacturers originally intended for the possibility of running at 2x60 and 2x66 MHz, but later ran into voltage threshold issues in the clock signal.