maxtherabbit wrote on 2021-06-15, 02:54:

As far as I know, the norm for ATX boards with 3.3v AGP is to just pass the power straight from the power header to the slot.

Perhaps that FIC board generates it's own 3.3v for some reason idk

Paadam wrote on 2021-06-15, 06:36:

The early AT boards with AGP slots and some early ATX boards (Asus P2L97 comes to mind specifically) had VRM's that converted 5v to 3.3v and they had problems with more powerful (for the time) graphics cards. Asus even had official rework guide regarding the P2L97 mod early revisions that included unsoldering VRM leg and routing wire from ATX 3.3v pin to to a capacitor leg instead.

Not aware that any of the BX boards had such issues.

The +3.3V thing has been an issue with lots of 1997-1999 AGP motherboards. With 440bx boards especially those by Gigabyte seem affected. Gigabyte fixed the issue by implementing a pair of jumpers in later designs that would feed the AGP slot 3.3V directly from the ATX connector.

I have only had Gigabyte 440ZX mATX board but that did not have any such problems (it had Tualeron 1200@1600, Ti4600 graphics etc) and it was my main gaming rig back in 2002-2005 after which I went to Asus P4GD1+CT479 and overclocked Dothan 😀

Doornkaat wrote on 2021-06-15, 12:18:

maxtherabbit wrote on 2021-06-15, 02:54:

As far as I know, the norm for ATX boards with 3.3v AGP is to just pass the power straight from the power header to the slot.

Perhaps that FIC board generates it's own 3.3v for some reason idk

Paadam wrote on 2021-06-15, 06:36:

The early AT boards with AGP slots and some early ATX boards (Asus P2L97 comes to mind specifically) had VRM's that converted 5v to 3.3v and they had problems with more powerful (for the time) graphics cards. Asus even had official rework guide regarding the P2L97 mod early revisions that included unsoldering VRM leg and routing wire from ATX 3.3v pin to to a capacitor leg instead.

Not aware that any of the BX boards had such issues.

The +3.3V thing has been an issue with lots of 1997-1999 AGP motherboards. With 440bx boards especially those by Gigabyte seem affected. Gigabyte fixed the issue by implementing a pair of jumpers in later designs that would feed the AGP slot 3.3V directly from the ATX connector.

I can understand it for AT boards (since the AT connector has no 3.3v obviously) but that was a really stupid design choice by Gigabyte for ATX boards.

Anyway sounds like a niche issue that affected a small handful of boards that has been blown out of proportion in internet lore

dionb wrote on 2021-06-13, 23:28:

Second this. I've had mixed results with different specimens of the same board, same revision (MS-6168 rev 2.0 is the first one […]
Show full quote

bloodem wrote on 2021-06-11, 10:02:

[...]

Yes, 440BX usually works fine at 133 MHz. But after you've thoroughly tested 10, 15, 20, 25 boards... at some point you'll encounter one that isn't 100% stable at this frequency (it could be either a chipset silicon lottery problem, or just poor build quality of that specific board).

Second this. I've had mixed results with different specimens of the same board, same revision (MS-6168 rev 2.0 is the first one that comes to mind 😉 )

Agree with Paadam though that it's as likely to be a board VRM issue as a fundamental chipset limitation. However you're generally stuck with a fixed combination of the two on any given board, so that distinction is fairly academic.

maxtherabbit wrote on 2021-06-13, 21:35:

A good way to determine if it really is the AGP overclock vs something else in the system would be to test with a PCI graphics card

It's a good indication, but not 100% unless the PCI card draws exactly the same amount of power as the AGP card does (V3-3000 AGP vs PCI would be a good test). Otherwise it could still be down to power issues.

I think we can rule out the PSU. I temporarily disconnected the 200W PSU I normally use in this system and plugged in an absurd Corsair RM750x, solely because I have it lying around unused. The crashes still happen with the machine running off of this.

Is there really any point in testing a PCI GPU? I could probably dig one out of storage, but I'm not sure it would tell us anything interesting.

pete8475 wrote on 2021-05-22, 15:46:

I've tried a number of cards in my CUBX-E machine, off the top of my head: […]
Show full quote

I've tried a number of cards in my CUBX-E machine, off the top of my head:

5900XT - perfect
6800XT - crashes
TI4200 - perfect
440MX - crashes
Voodoo 3 - perfect

I'm sure I've used others but that's what I can remember for sure right now.

Not all Voodoo3 models will work at 89Mhz. I tried mine last week and it crashed at 89Mhz but worked perfectly fine at 66Mhz.

I have another model to test though.

mockingbird wrote on 2021-06-25, 12:34:

Not all Voodoo3 models will work at 89Mhz. I tried mine last week and it crashed at 89Mhz but worked perfectly fine at 66Mhz.
I have another model to test though.

In my experience most of them work (in fact I now have 5 x Voodoo 3 3000 AGP, one of them which I just received and tested today), and they all work on 440BX @ FSB 133 MHz. However, as weird as it sounds, it also seems to be motherboard dependent. For example, I usually test them on an Abit BH6 board where they work fine at any FSB (and AGP) frequency, but at some point I tested one on an Amptron 650B-ATX and it did not work. On the other hand, on the same Amptron board, nVIDIA cards work perfectly at any FSB setting.