I imagine that not all 440BX chips accept a 33% overclock. Like any other IC.

But yes definitely try another video card as there's no guarantee with them either.

Yes,
I put in the following AGP graphics cards, and tried them at 133MHz FSB, 89MHz AGP:
-Voodoo 3 2000 16MB
-Generic Geforce 2 MX 32MB
-Asus TNT2 Vanta 16MB
-Sparkle TNT2 M64 32MB
-MSI Geforce 6200A 256MB
-Sparkle Geforce 2 MX 64MB

Results vary, but all combinations lock-up within 10 minutes... Except with that last geforce MX the system kept running! Did 40 minutes of DOS gaming (System Shock CD) and a little windows. No problems...

So I am starting to suspect 89MHz AGP tolerance differs alot between very similar chips, and the same goes for a Voodoo 3 vs another Voodoo 3.

I've never really had a rock solid 89mhz AGP experience. Cards will seem stable but they eventually freeze in 3D games. Maybe it takes an hour. Not many people test extended use.

And yeah I'm sure every video card varies even in the same make and model because you are going beyond what they were validated to do.

If you can set a divider from the bios then a PCI card would be the way to go. I think its the FSB:PCI that will run 1/3 or 2/3.

RogueTrip2012 wrote:

If you can set a divider from the bios then a PCI card would be the way to go. I think its the FSB:PCI that will run 1/3 or 2/3.

Iirc BX can be set with PCI divider 4, making a 133Mhz BX overclock only it's AGP port.

swaaye wrote:

I've never really had a rock solid 89mhz AGP experience. Cards will seem stable but they eventually freeze in 3D games. Maybe it takes an hour. Not many people test extended use.

I would say so too. Even that 40 minutes of running on that last Geforce MX is insufficient to approve it. It is just to get a better understanding. Normally I would not bother with 89 MHz AGP, but I understood any Voodoo 3 could handle it, which I now know is not true.

Also I kinda liked the idea of changing a 200 MHz Pentium into a 400 MHz Pentium with the convenience of SoftFSB for windows.

Tetrium wrote:

IIRC BX can be set with PCI divider 4, making a 133Mhz BX overclock only it's AGP port.

The Proper 133 MHz/4 PCI divider is present in the i440BX. So at that speed only AGP is the trouble spot, the rest is OK. Or maybe the i440BX Chipset itself may get unreliable. But so far I would say it is just the AGP complaining.

Edit: just managed to lock up the system with the Sparkle Geforce 2 MX 64MB at 89MHz. And again with another Geforce 2 MX 64MB. Humble Battle Isle 2 sure brings out the worst of the videocards, maybe because of its non-standard video mode. I just ordered a Voodoo 3 PCI and will write about my results as soon as I have the chance to test it. Hmm 18 Pounds excl shipping.... well it is for a good cause.

I am really interested in this thread. I have set up an Asus P2B rev 1.12 mobo. It uses the i440BX chipset. I have set the bus speed to 133 MHz speed. I am using a 1.4GHz 133FSB Tualatin server chip, in a PowerLeap adapter (revision 2). I've got 2 sticks of 128MB CL2 SDRAM (total 256MB) in it, and I'm using a Voodoo3 2000 AGP VGA. I decided to run these tests in this order

1) I ran SuperPi, and selected the longest test available which was 32M. I ran this test, and let it continue in the background...
2) While SuperPi was running, I ran Prime95 and selected the default test which is called "Blend". I ran this test, and let it continue in the background...
3) While both SuperPi and Prime95 were running, I ran WinTune 98, and set it to run its tests 9 times. I unchecked the Direct 3D test option, because it complains about not finding a Direct 3D device.
4) After WinTune 98 had finished running its 9 tests, I ran WinBench 99. (I did not close WinTune 98.) I selected "custom tests", and choose all of the graphics tests. There were 47 of them.
5) After the 47 WinBench 99 tests had finished, I ran Quake2 (I did not close WinBench 99), and ran the Quake2 demo called timedemo demo1.dm2. I let it run for about 7 minutes. It ran it in 3dfx OpenGL mode, with a resolution of 1024x768. Afterwards, I quit Quake2.
6) I then ran 3DMark 99 Max's default benchmark test twice. Afterwards, I did not close 3DMark 99 Max.
7) I then repeated all tests from 3) to 6) inclusive.

Test start time = 13:39 PM
Test end time = 15:24 PM
Total running time = about 1 hour and 45 minutes.

The overall result: no crashes, no lock ups, and no strange stuff. I remember testing an AOpen AX6BC, and finding it curiously unstable. I think my tests were set to 133 MHz on that board. I do remember getting lock ups. Gerwin, I think the problem is not i440BX + 89MHz AGP = problem, but AOpen AX6BC + 89MHz AGP = problem. I think you need to test a different mobo. The Asus P2Bs are about the best you can get. You even get 3 ISA slots! 😀 I've tested this thing @ 150 FSB, and never had a lock up. I might run those 1-7 tests again, at 150 FSB. Time permitting...

Just one last thing. I find this fascinating, but after running these tests for about 1.75 hours, I could very comfortably rest my fingers on the V3's heatsink (while the tests were still running), and not feel any discomfort. And this particular V3 is the one with the "thin/not tall" heatsink, that weighs 10 grams less than a regular AGP V3. Weird! Having said that, the mobo is not inside a case. Perhaps that makes a huge difference?

Edit: On the mobo, there is a jumper called AGPFS. It is set to 1-2. The other jumper option is 2-3. I haven't got the P2B manual to hand right now. I wonder if this setting is making the AGP port run at a safer speed? It must be some kind of AGP "divider". Does anyone know what this jumper does, for the Asus P2B revision 1.12 mobo? Thanks.

Edit 2: I can answer my own question. I found the relevant information about the AGPFS jumper, from one of my previous posts! I have copied and pasted this info below. It means that the AGP bus is currently operating at a speed of 133 * 2/3 = 89 MHz.

The 3 pin PCIRATIO jumper has been completely "soldered out", meaning that you cannot put jumpers on it.
The 3 pin AGPFS jumper is set to pins 1-2 = "on". The settings for AGPFS are -

1-2 AGPCLK = CPUCLK * 2/3 (Currently Enabled = "on")
2-3 AGPCLK = CPUCLK

retro games 100 wrote:

...and I'm using a Voodoo3 2000 AGP VGA. I decided to run these tests in this order

...
The overall result: no crashes, no lock ups, and no strange stuff.

Gahhh!!...and only yesterday I found out I have only 3 V3 AGP cards total! Time to quickly get a couple cheap before Ebay sends their prices skyrocketing also 😜

retro games 100 wrote:

Just one last thing. I find this fascinating, but after running these tests for about 1.75 hours, I could very comfortably rest my fingers on the V3's heatsink (while the tests were still running), and not feel any discomfort. And this particular V3 is the one with the "thin/not tall" heatsink, that weighs 10 grams less than a regular AGP V3. Weird! Having said that, the mobo is not inside a case. Perhaps that makes a huge difference?

This definitely makes a difference. In a badly build system the V3 will not receive enough air circulation, with the hot air just hanging around the graphics card.
Btw, the only 2 different coolers that I know of are those quare (usually black) ones with the same dimentions as the GPU itself, and the larger Aluminium heatsink that kinda overhangs part of the pcb on the left and right side.
Generally the ones with the smaller heatsinks are V3 2000's and the ones with the larger Alu heatsink are V3 3000's 😉.

Not much time to test and write. But I found a jumper concerning the AGP divider and set it from "auto" to "2/3 manual" and the voodoo 3 behaves better so far. I think the auto setting is kinda dumb in that it only detects the CPU FSB rating and not the actual FSB.

I set the P2B mobo's FSB to 150, which means the AGP is running at 150*2/3= 100 MHz. When the desktop appeared, the system appeared to have locked up. I wasted no time in testing this situation any further, and so I uninstalled the V3 card. In its place, I installed a GF3 Ti200. That works! I ran through all of the tests numbered 1-7 again. No problems! I decided not to run all of these tests twice, because I didn't want to spend that much time on this.

One thing about these new tests that was interesting - I set the Powerleap's voltage from 1.45 (default for the Tualatin) to 1.65. I did this because the 1.4GHz CPU needed a boost, as it was running at 150 FSB, as opposed to 133. Usually, 1.6 volts is OK for this kind of boost, but I was lazy and the jumper setting for 1.65 was easier to set. However, after 22 minutes of testing, I noticed that Prime95 had failed. So I increased the voltage from 1.65 to 1.7, and that fixed the problem. I reran all of the tests 1 to 7 again, and no problems occured.

BTW, Tetrium, I have got 2 different Voodoo3 2000 AGP cards. They both look identical, except for the heatsink. It looks the same, but if you look very carefully, one heatsink is slightly taller than the other heatsink. The cards weigh different amounts. The one with the shorter heatsink weighs 10 grams less than the other card.

retro games 100 wrote:

BTW, Tetrium, I have got 2 different Voodoo3 2000 AGP cards. They both look identical, except for the heatsink. It looks the same, but if you look very carefully, one heatsink is slightly taller than the other heatsink. The cards weigh different amounts. The one with the shorter heatsink weighs 10 grams less than the other card.

Ah ok. That's quite interesting. I never really bothered to check for minute differences in standard heatsinks of otherwise statistically identical cards (though otoh, the 3 V3's I own are all the 3000 variant. I wanted to build a Nehemiah rig with a V3 2000 AGP, only to find out yesterday that I don't own one 😵 ).

Also, it's not just the weight that's important, it's also how the heatsink is built.
*Lots of fins/"pins" > Few fins/"pins"
*Larger/wider/taller > Smaller/slimmer (obviously 😜)
*Thicker base > thinner base
*Copper > Aluminium > most other metals
And last but not least, how's the TIM? Is it a thick layer? Is it missing, or is it spread thinly and evenly?
(Lol, I could write a mini-guide about this subject, along with pics etc 😜)

A heatsink can be totally awesome, but that will be useless if a chip can't transfer it's excess heat to the heatsink due to a shoddy layer of TIM.
And a heatsink without airflow is kinda useless as well 😉
And the ability of a heatsink to remove excess heat from the chip it is cooling also depends on the ambient temperature...or to be exact, the greater the discrepancy between the temperature of the heatsink and the surrounding air, the more effective the heatsink can get rid of it's collected heat 😉

Tetrium wrote:

(though otoh, the 3 V3's I own are all the 3000 variant. I wanted to build a Nehemiah rig with a V3 2000 AGP, only to find out yesterday that I don't own one 😵 ).

There is no problem in that, just adjust the BIOS so it runs at the V3 2000 speed of 143MHz. load it to the EPROM and reboot.

RG100 wrote:

I think you need to test a different mobo. The Asus P2Bs are about the best you can get. You even get 3 ISA slots!

Well P2B is also the first generation of BX mobo's. No jumperless config. And P2Bs appear in quite a few revisions, all with their peculiar things.

Anyway, it is possible again that the blame does not go to the mobo, as I might have been using no 89MHz AGP, but 133MHz!

One question: Does your P2B do 50MHz FSB? (all three jumpers in position 2-3)

gerwin wrote:

Tetrium wrote:

(though otoh, the 3 V3's I own are all the 3000 variant. I wanted to build a Nehemiah rig with a V3 2000 AGP, only to find out yesterday that I don't own one 😵 ).

There is no problem in that, just adjust the BIOS so it runs at the V3 2000 speed of 143MHz. load it to the EPROM and reboot.

Thanks, but eh...I am a complete newb when it comes to this kind of things

gerwin wrote:

Anyway, it is possible again that the blame does not go to the mobo, as I might have been using no 89MHz AGP, but 133MHz!

🤣 Perhaps that is why my tests were failing on this mobo, too?

gerwin wrote:

One question: Does your P2B do 50MHz FSB? (all three jumpers in position 2-3)

The FSB information printed on the PCB says that 66 is the lowest setting. Pity! Time permitting, I may experiment and look for undocumented settings...

Tetrium wrote:

how's the TIM? [on the Voodoo3 card] Is it a thick layer? Is it missing, or is it spread thinly and evenly?

It looks fairly thin and evenly spread.

Tetrium wrote:

Thanks, but eh...I am a complete newb when it comes to this kind of things

From the Voodoo 3 Topic, one fills in the corresponding K,M,N values in tdfxbioseditor_1.61.exe, then apply and save... and flash:

gerwin wrote:

As I cannot find a clock setting utility for Dos (who has one?), I might just edit the BIOS of the card with tdfxbioseditor_1.61 […]
Show full quote

As I cannot find a clock setting utility for Dos (who has one?), I might just edit the BIOS of the card with tdfxbioseditor_1.61.exe so it runs like a Voodoo 3 1000.

Copy code to clipboard

119-04-2011 GB, INFO FROM BIOS FILES:
2VOODOO 3 PLLCtrl
3TYPE   1000       2000      3000      3500     3000/O.C.
4MHZ    125,090    143,182   166,091   183,750  200,455
5K      1          1         1         1        1
6M      11         1         3         1        3
7N      225        58        114       75       138

retro games 100 wrote:

gerwin wrote:

One question: Does your P2B do 50MHz FSB? (all three jumpers in position 2-3)

The FSB information printed on the PCB says that 66 is the lowest setting. Pity! Time permitting, I may experiment and look for undocumented settings...

Nevermind, I found that only the earliest P2B's have a 50MHz to 133MHz FSB range. And since yours does 150MHz it will have to be a later one.

I extracted and combined some info from SoftFSB and Sisoft Sandra, so I now have this list of at least 60 Mainboards and their exact possible FSB settings. 😀

I decided to order for another Pentium II/III 440BX mobo. A Gigabyte GA-6BXC Revision 2.0. Some people here at vogons have it too. I think I may prefer it compared to other BX boards, because...
- 3 Isa slots as the earliest BX boards used to have, compared to the usual 2 or even 1 or 0 in later BX's.
- Rev 2.0 Has PC99 colored connectors, so it is a late production BX.
- Rev 2.0 supports Coppermine P-III CPU's.
- The jumpers are less easy to operate, but they do give me full control of the connections. SoftFSB will still work to override the jumpers, I hope.
- It has 66..133MHz FSB settings . But I know that the thing unofficially has a 50MHz FSB setting, like the earliest Asus P2B's. So I want to try my P-III at an even lower speed: from 200MHz to 150MHz. Or 100Mhz in the case of a multiplier limited P-II. 😀

I read 5u3 had a problem with the ACPI steeling an IRQ on this board. I hope that won't annoy me.

Gerwin, this looks promising! 😀 I think I've got a Gigabyte GA-6BXC revision 2.0 somewhere. Please can you tell me what the unofficial 50MHz FSB setting is? Thanks a lot.

Yes you wrote about your board on vogons before, it is the same.

It has a PLL chip supporting eight preset frequencies, no more. The 4 dip switches give 16 combinations.
Google told me one should try On/On/On/Off or On/On/On/On for 50Mhz FSB.