Congratulations on your two Biostar version 1 mobo enhancements! Good work. 😀 I'm sorry, but I can't answer your question, but I do have a suggestion. There's been quite a bit of discussion over the last several months about OC'ing 486s to 200MHz. Over this time period, ebay "486 sellers" have been enjoying themselves somewhat. 😉 My guess is that plenty of people by now have tried the 200MHz overclock test. I'm intrigued why the success rate seems so low - that is, to produce a stable system with accompanying benchmark screenshots. That leads me to wonder...

What PSU are you using? I wonder if this is important. I'm using a fairly high spec (~450W) ATX PSU, with an ATX to AT adapter. I wonder if many 486 users are quite naturally using old 'n' crusty AT PSUs, and these PSUs aren't good enough for this type of "mad science" experiment.

I suspect the 200 MHz overclock success rate is so low because there are few of the AMD X5-133ADW CPUs w/CPUID=04F4, or the rare X5-160ADZ. I am using an ancient 1997 PSU for these particular tests, but the 180 MHz test worked fine.

I doubt the culprit is the PSU. Even if the PSU isn't regulating the AC-to-DC voltage well, that is to say, there are surges or noise, the on-board motherboard voltage regulator should take care of that for 3.3 and 4V. However, for 5V, I'd have to look at the MB circuit to see if the motherboard voltage regulator is being bypassed and the 5V is being taken directly from the supply. If it is being bypassed, I suppose I could hook up an analogue multimeter and watch the voltage for spikes and noise. An oscilliscope would be better, but I don't have one at home.

That aside, I may have discovered a bigger problem with this Ver.1 board. Aside from the fussy cache timing requirements using 512KB cache, as mentioned above, it seems that the PCI BUS cannot be run at 33 MHz or higher. When using a 33 MHz FSB CPU, i.e. AMD X5-133/4X or Cyrix 5x86-100/3X, Windows won't boot when the FSB-to-PCI divisor is 1:1, but runs fine at 1:2/3. I've tried every possible cache and memory setting, but nothing lets me boot into Windows with 1:1. I suppose I could try running a FSB of 25 MHz and just converim that 1:1 works, that is, something isn't messed up in the BIOS.

Considering that I was able to run the AMD X5-180 at 60x3 using a 1:1/2 PCI divisor, so I'm guessing that a 30 MHz PCI BUS is ok, as well as a 27 MHz PCI BUS (40 MHz x 1:2/3).

What gives? I'm at a loss here. Is the Northbridge memory controller gone bad? I also tried 3 different BIOS versions, including the original, but nothing seems to help. I am currently accepting any ideas, even the crazy ones.

My guess it also depends on the age of the silicon. The younger the chip is, the better it's quality was when it was produced, similar to how newer Bartons generally overclock better then the earlier ones.

An old PCI bus that cannot work at 33 MHz may be the result of the growth of crystalline impurities within the chipset(s). In otherwords, a rotten lemon. I'm sure at one point it did work at 33 MHz, as this was the PCI spec, but this boy is past his prime.

feipoa wrote:

An old PCI bus that cannot work at 33 MHz may be the result of the growth of crystalline impurities within the chipset(s). In otherwords, a rotten lemon. I'm sure at one point it did work at 33 MHz, as this was the PCI spec, but this boy is past his prime.

Have you tried using a DX-33 or DX2-66 chip in it?

feipoa wrote:

An old PCI bus that cannot work at 33 MHz may be the result of the growth of crystalline impurities within the chipset(s). In otherwords, a rotten lemon. I'm sure at one point it did work at 33 MHz, as this was the PCI spec, but this boy is past his prime.

Now this is a legit argument, forgot about the crystals 😉
Btw, I'd say it's logical to assume that any previous abuse of any silicon by a previous owner could also lower the max stable frequency at which a chip can run stably

I didn't try a DX33 or DX2-66 as I'd never run such a slow CPU in the system. I'm really just looking for a backup motherboard to my current MB8433-UUD v3.0.

Agreed, you never know what crazy abuse any of this eBay IC stuff has undergone.

Big warning to Biostar MB8433-UUD owners,

if you clock your FSB to 66 MHz, and later try to go back to 33 MHz, you may no longer be able to operate your PCI bus at 33 MHz. It seems something has gone screwy with the 1:1 BIOS option for the HOST-to-PCI BUS setting.

You can continue to operate at 66 MHz FSB and use the 1:1/2 divisor option, or operate at 40/33 MHz and use the 1:2/3 option, but 1:1 is out of the question.

This is not an isolated case, three (3) MB8433-UUD's are acting like this after running at 66 MHz FSB. One would logically think that one of the PCI cards must have gone bad. I have identical backup PCI cards to the 3 that are in the system, and those didn't fix the problem.

I can sometimes boot into Windows at 1:1 and 33 MHz, but it will crash. 1:2/3 is forever stable. Changing CPU's didn't fix the problem. I tried AMD X5's and other Cyrix 5x86 CPUs, but no dice. MemTest is all clear, the HD checks out. The funny thing is, 66 MHz FSB with 1:1/2 for the PCI works fine with a Cyrix 5x86-120GP at 133 MHz. What gives?

It seems something must have gone bad in the chipset, where I figure the 1:1/2, 1:2/3, and 1:1 operations take place. I've tried different BIOS chips, but that didn't fix it. I probed the PCI bus with my oscilliscope at 1:1 and jumpers set at 33 MHz, the waveform is clean at 33.33 MHz.

The possible fix? I have ordered some clock dividers from digikey. My plan is to continue running the FSB at 66 MHz and implement the 1:1/2 setting. There's a spot just before the CLK signal reaches the CPU where I can easily put in a 1/2 clock divisor chip (from digikey). This will bring the CLK back to 33 MHz as seen by the CPU so I can use 3x and 4x CPUs again.

I've tried replacing RAM, cache, BIOS defaults, putting RAM in different slots, use different types of RAM, removed all PCI cards except graphics and SCSI, changed PCI IRQ's in the BIOS, etc. Nothing works unless I put the PCI BUS back to a 1:2/3 divider. But who wants to run their PCI bus at 22.22 MHz? Ugh.

You've been warned! I'm at a loss. Anyone experience this before? Your wild ideas are welcome!

I've got an idea, but it really is a wild guess. You said that they're welcome. 😉 I see from your post that you have tried virtually everything. How about making an exhaustive list of all the possible bus jumper speed settings. Then, try each setting that does not configure to a known speed.

My wild guess is that one of them may "reset" the way that the chipset sees what bus speed you really want. It's almost as if the mobo / chipset seems to be trapped in to thinking you want the board run at 66, and can no longer see that you've changed the jumpers back to 33.

Shock the chipset into submission 'eh? Its a wildcard, but I kinda like it. I've added it to my brainstorm list. I've already probed each of the 14 pins of the motherboard's PLL multiplier FSB chip. If anyone wants to know what each pin is for, let me know.

I appreciate your idea, thank you. People, keep-em coming!

It is almost as if the PLL which handles the 1:1, 1:2/3, 1:1/2 is no longer synchronized to the rest of the system at 1:1, as if there is a slight phase shift, or perhaps just enough drop in voltage and/or duty cycle that the system cannot handle it. The scope still sees 33.33 MHz with voltages similar to that on a typical PIII board, so it seems that the chipset itself just doensn't like 1:1, as if something burned up a bit and you need a higher frequency to "get though".

The funny thing is that the system worked perfectly before trying 66 MHz. I even backed up my HD from a good cloned state, but that didn't help. Most of the BSOD errors are related to an incorrect checksum image, hard disk error, inaccessable boot device etc, or Win32 something driver. When I run the system from the built-in IDE port it runs fine at 1:1, so I wonder if the built-in IDE port bypasses this 1:1 deal some how?

I ordered another SCSI PCI card just incase both of mine are acting up. Not likely, but it was only $2+8. As another long-shot test, I'll try a different RTC and try a clean insteall on a different harddrive. Maybe the controller chip on the SCSI harddrive went bad? I've tried 4 different LVD-terminated cables, but no go.

Right now my highest hopes are for the 1/2 custom CPU clock dividor. If it works, you get to keep the 66 MHZ speed to the RAM/cache and the 33 MHz for the PCI bus. I am starting to question the long-term stability at 66 MHz inasmuch as the chipset is concerned. The PLL chip on the motherboard, which is the same as other UMC based boards, may have just added the 60/66 MHz option in hopes that their chip would also sell to the socket 5/7 market. While the UMC 8881/8886 chipset may be able to handle 66 MHz operation, the PLL divider circuit inside the chipset may not.

I've looked at all the other chips on the motherboard, and I do not see a PLL divider chip (only the main PLL multipler chip for the FSB). So I am assuming that the PLL division occures in the UMC southbridge chipset. Can a BIOS contain a clock divider?

Very puzzling. I think working for a MB maker would be kinda fun.

Fascinating post! So the (Edit: PCI-based End Edit) built-in IDE port works at 1:1. My guess is that an ISA-based I/O controller would work too. I wonder what would happen if you tested your boards with a PCI-based IDE controller card? That's just another wild guess idea - simply to tick that test off an exhaustive list of tests to do.

feipoa wrote:

Maybe the controller chip on the SCSI harddrive went bad?

I read your list of test replacement parts above, but I couldn't spot a HDD. Do you have a spare?

I am pretty sure that the IDE controller is PCI-based since it supports PIO-4. There's no way to get that fast on the ISA bus. I do have an ATA PCI card I can test out. Its been added to the "if everything else fails" list.

I've been running the system for most of the day at 2x60 MHz on a Cyrix 5x86-120. I'm using a 1/2 FSB-to-PCI divider so the PCI bus is at 30 MHz. I haven't had any problems with this setup what so ever. I even transfered 65 GB of data over the network without a hickup.

So I am pretty sure it isn't anything SCSI harddrive related, but somehow the IDE controller on the motherboard bypasses the crash mode of the 1:1 divider. Outside of the M919 relm, this is by far one of the most bizaire experiences I've had with a 486.

Say, didn't you overclock your Biostar to 60/66 MHz? At a time which is most convenient for you, could you run an AMD at 4x33 MHz with a 1:1 ratio, not using the onboard IDE port? Preferably in NT 4.0? Bah, nevermind, this is too specific of a request. Anyway, I find that if it can reboot 3x in a row without a crash, then the system should be good. During my tests, sometimes it crashed on reboot #1, 2, or 3, but if it passes 3, it doesn't crash on 10x more reboots.

The good news about running at 2x60 MHz is that the cache/memory are quite fast, but I'm loosing 13.3 MHz on the CPU clock. The other good piece of news is that branch prediction (BTB) works at 120 MHz (it doesn't work very well at 133 MHz due to heat).

Now, from the combined results of the Ultimate 486 Benchmark Comparison and the Cyrix 5x86 Register Enhancements, we can theoretically compare the differences between a Cyrix 5x86-133 (4x33 without BTB) and a Cyrix 5x86-120 (2x66 with BTB).

120 MHz CPU ----> 133 MHz CPU (loss of 13.3 MHz)
Drops performance by 8%

33 MHz FSB ----> 60 MHz FSB
Increases performance by 5%
(well, I have results for 66 MHz, which is 6%, so I dropped it 1% for 60 MHz)

no BTB ----> BTB (with branch predition)
Increases performance by 8.5%

33.3 MHz PCI BUS ----> 30.0 MHz PCI BUS
Drops overall system performance (ethernet, harddisk, graphics) by 2%?

The end result is that the system is still equally as fast as a Cyrix 5x86-133, though it looses its prestige somewhat. Unfortunately, my stock Cyrix 5x86-133 didn't work longer than a 20 min. with BTB on, but it will if you leave the case off. Perhaps some stock Cyrix 5x86-133's can operate warmer with BTB on.

Thanks for correcting my mistake about the IDE controller. I have edited my post above containing this mistake.

Yes, I did overclock my Biostar mobo to 60 and 66. I wonder if I have damaged an aspect of the mobo, without knowing. The reason why I say this is because I have not tested any PCI-based IO cards with this mobo. I only use ISA-based IO cards, and sometimes the onboard IDE connector. At the moment, I don't have any SCSI equipment, and I don't own a copy of NT.

I'm 99% sure that after my 60 and 66 overclocking experiments, I tried setting the FSB back to 33MHz with 1:1. The reason I am confident that I tried this, is because that's my "is everything still working?" test, which I always do after any overclocking.

At the moment, it's very difficult to do any testing, because my equipment is at a different location. I hope this will get resolved next year, and I can start testing again. On my list of testing jobs to do is - find a PCI-based IO card (it'll have to be IDE-based), and try 33 FSB with 1:1.

I see some licensed Windows NT disks on eBay for cheap (~$7).

If you decide against the SCSI route, I highly recommend the Promise SATA150 TX2plus for retro PCI IDE controllers. It has 2 SATA ports and 1 IDE ATA/133 port. What is better is that it is I've tested it out personally. I only tested the IDE port on it. It is rare for a SATA card to have NT 4.0 drivers, but this unit does.

If anyone is interested in how branch prediction works,
http://igoro.com/archive/fast-and-slow-if-sta … ern-processors/

After some troubleshooting, I now know all the possible FSB frequencies of these UMC chipset motherboards. Refering to the Biostar in particlar,
it can run at 20, 25, 33, 40, 50, 60, 66, and 83 MHz. For 20 MHz, JP15, JP16, JP17 all jumpers off. For 83 MHz, JP15=ON, JP16=ON, JP17=OFF. At 83.33 MHz, the CLK signal is pretty weak. It is also pretty weak at 60 MHz, but things seem to work fine.

Does anyone know what the maximum acceptable clock voltages are for 486 CPUs and motherboards?

The Intel 486 Embedded Developers Manual provides a typical clock diagram listing 0.8 V as the cut-off low state, 2.0 V as the high state, 1.5 V as the reference (probably a comparitor ref.), and 2.5 V as the peak voltage of the clock waveform. It, however, also mentions that "CLK only needs TTL levels for proper operation." TTL levels state a low is 0-0.8 V and a high is 2.2-5.0 V

On the other hand, the Intel 486 Embedded Hardware Reference, pg 10-31 mentions that a low can be from -0.3-0.8 V and a high from 2.0-(Vcc+0.3V). So, if the chip is run at 3.45 V, that high range would be 2.0-3.75 V. Add an appropriately sized resistor(s) just before the PCI CLK signal?

Note that these are the clock (CLK) voltages going into the CPU from the motherboard. What I do not understand, is how the system runs correctly, even at 33 MHz. At 33.33 MHz FSB, I get a Vmin=1.06V and a Vmax=2.50V. According to the Intel low state, this CPU never has a low clock. What is even more shocking, is that at 60 MHz, Vmin=1.65V and Vmax=1.95V. In this case, there is neither a high nor a low, yet the system works perfectly. So there must be some conditioning, perhaps amplification, done by the CPU's onboard PLL.

I may now be getting some insight for why the MB8433-UUD no longer works with a 1:1 FSB-to-PCI divider at 33 MHz. There may not be any conditioning done by the chipset. at 1:1, I get a PCI Vmin=1.04V, Vmax=2.60V. At 1:2/3, I get a PCI Vmin=0.8V, Vmax=2.05V.

At 1:1, the PCI clock voltage doesn't drop to the level needed to see a low clock, but when using the 2/3 divider, it does. So to fix the motherboard, it would seem that we need a DC bias drop of ~0.3 V.

Isn't 1.5x reserved for 4x on AMD 5x86's? That would kinda limit the usefulness of the 83mhz setting. Pushing a 133mhz chip to 166 using 83 x 2.0 will probably only result in a dead chip, and I'm sure nobody with a working 150 or 160mhz model is going to risk it. It makes me wonder why they would even include that setting on a 486 motherboard. 60 and 66 are unusual enough as it is. Surely no 486/5x86 could ever use an 83mhz bus. The ISA and PCI slots would probably be way out of spec at that speed, too.

I am speculating, but I think that the makers of the UM9515 PLL (the chip that is on the motherboard which sets the various FSBs from a single 14.34 MHz crystal) were hoping it could also be used in socket 5/7 motherboard.

This begs the question, why isn't there a 75 MHz setting then, instead of an 83 MHz setting? It could be that there was speculation of an 83 MHz setting for the Pentiums, or that it was too difficult or costly for them to integrate a PLL multiplier on the chip for 75 MHz, and 83 MHz was more easily obtainable. I am only speculating here. This question is best answered by someone who worked in the motherboard creation industry in the mid-90's.

If anyone has any wild guesses about the CLK voltage issue mentioned above, please don't hold back. I have four MB8433-UUD's which need fixing.

feipoa wrote:

I ordered another SCSI PCI card just incase both of mine are acting up.

+

feipoa wrote:

... Maybe the controller chip on the SCSI harddrive went bad?

+
the idea of trying a PCI IDE-based HDD controller.

Have you had a chance to try any additional HDD related components on these mobo(s)?

Edit: Sorry, added in "IDE" to 3rd line.

As a matter of fact, I just started working on this again today. I got my $2 replacement SCSI card from Isreal and I am using a MB8433-UUD which was never overclocked.

I am having the same troubles: I need to use a 1:2/3 divider for the PCI bus. I'm going to throw in an Ultra160 SCSI card to see if this helps. I am pretty sure the issue is related to the SCSI card as most of the errors are harddrive related. It just doesn't make any sense how I was able to use it fine for many years and now it won't work at 33 MHz anymore.

Trying the ATA133 PCI controller card will be my final step because I'm not really interested in running a non-SCSI system long-term, but those results will help with a SCSI-only conclusion.

After the U160 controller, i'm going to try a non-terminated SCSI cable and terminate on the harddrive with the jumper. If all of this still fails, I'll go back my EE lab and try to redesign the motherboard.

My plan is to run the system at 66 MHz and use a 1/2 PCI divider, then I want to add in a custom CPU clock multiplier of 1/2, match the line impedance, and voila! The problem with this is that the clock divider (1/20 I got needs more definatition between the high and log of the clock signal for it to work. It didn't work as it stands; I need to add a pre-amplifier to the clock signal to bring up the voltage levels, then divide by 2.

As a backup to this plan, I'm ordering all new 1997 era chipsets for these motherboards, including the northbridge, southbridge, superI/O, PLL, and serial chip. I already tried replacing the clock driver chip, but that didn't help.

So those are my 2 backup plans. I suppose I should try the PCI-ATA133 card before those two plans, but I at least want to get the PO out for the new chipsets.

I just as well add to the list is testing this in an M919. It has the same chipsets as the MB8433 so there shouldn't be too much problem with that.

EDIT: I seem to be having some initial success with setting the max SCSI speed to 20 MB/s. 10x reboot successful w/1:1 divider. I'm now going to increase it to 40 MB/s and see what happens. Before I had it on 160 MB/s and 80 MB/s.

EDIT: From Adaptec's SCSI Bench, the max speed I can get on the 486's PCI BUS is 31 MB/s at 33 MHz, irregardless of the SCSI card's speed setting. 40, 53, 80, and 160 MB settings all yield 31 MB/s with SCSI Bench. 20, 27, 32 MB/s settings in the SCSI BIOS all yield 19 MB/s with SCSI Bench. The 40 MB/s setting boots more times consecutively than the 80 MB/s, but I can't get 10 consecutive reboots without a BSOD. So it would seem that the success of the 20 MB/s settings can be attributed to how much data can be shoved onto the 486's PCI bus at one time. Fishy is that it worked fine before, and it works on the 80 MB/s with 66 MHz FSB divided by 2.

EDIT: The PCI clock voltages are different when using a 1:1 and a 1:1/2. This is the only reason I can think of as to why I can use the Adaptec card on 80 MB/s using 1:1/2 (33 MHz) and only up to 20 MB/s using 1:1 (33 MHz). The faster operation needs a more defined PCI clock which goes well below 0.8 V for the a LOW and well above 2.0 V for a HIGH. It could also be related to how much noise is on the CLK using the different divisors. If there is too much noise and not enough comparitor hysteresis is added to the card's referance voltage detection, you'll count more clocks than there actually are.

EDIT: I wonder if the issue can be ASPI related?

EDIT: I tested out the ATA133 PCI card by first installing the drivers onto the CF card using the onboard IDE port in NT4. I then switched the CF card onto the ATA133 card. The system booted, but after typing in my Windows NT password, it freezes at "logging on". I waited 3 min., but no change. Perhaps it is necessary to install NT from scratch when using this ATA133 card? My method of installing the driver first worked when switching from Ultra2 to Ultra160 SCSI cards, but not for this ATA133 card.

EDIT: For those of you who are still using a CF card on your onboard IDE port (486), take a look at these benchmarks,

Ultra SCSI hard drive: 32 MB/s
CF Card PIO4: 2 MB/s