That's typical for early Via chipsets. Both my Tyan S1854 (Apollo Pro 133A) and Soyo 7VMA-B (PM133) have this issue, and I just disable 4x in the CMOS settings.

Repo Man11 wrote on 2026-08-12, 14:53:

That's typical for early Via chipsets. Both my Tyan S1854 (Apollo Pro 133A) and Soyo 7VMA-B (PM133) have this issue, and I just disable 4x in the CMOS settings.

But it's not happening just at 4x, but at 2x too, or is that also typical? Being limited to 2x probably wouldn't matter much for performance, but I wonder if 1x starts to somewhat affect it.

Could be bad caps.

It's not uncommon to see NVidia drivers force AGP 1x on non-Intel AGP 4x chipsets though. It's hard to say if you might notice a bottleneck. It would probably be more substantial with games that you wouldn't run on GeForce FX anyway.

Mike_ wrote on 2026-08-12, 18:36:

Repo Man11 wrote on 2026-08-12, 14:53:

That's typical for early Via chipsets. Both my Tyan S1854 (Apollo Pro 133A) and Soyo 7VMA-B (PM133) have this issue, and I just disable 4x in the CMOS settings.

But it's not happening just at 4x, but at 2x too, or is that also typical? Being limited to 2x probably wouldn't matter much for performance, but I wonder if 1x starts to somewhat affect it.

Neither of my Socket 370 boards has the option of 2x - it's either 4x or 1x, and 4x has been hopelessly unstable with Nvidia cards every time I've tried it.

With my Soyo K7V (KT266 which has no issue with AGP4x) and a Ti 4600: 3D 2001 scores are 4x 12,381, 2X 12,228 1x 11,443 so there is a penalty.

Try experimenting with turning Fast Writes off if they are enabled.

swaaye wrote on 2026-08-12, 19:33:

Could be bad caps.

I have recapped the motherboard, and the GPUs seem to have quality caps so that's unlikely.

Repo Man11 wrote on 2026-08-12, 19:42:

Neither of my Socket 370 boards has the option of 2x - it's either 4x or 1x, and 4x has been hopelessly unstable with Nvidia cards every time I've tried it.

With my Soyo K7V (KT266 which has no issue with AGP4x) and a Ti 4600: 3D 2001 scores are 4x 12,381, 2X 12,228 1x 11,443 so there is a penalty.

Try experimenting with turning Fast Writes off if they are enabled.

I have already tried turning fast writes off, and it had no effect.

Anyways, it sounds like 4x doesn't matter but it would be nice to get it to work at at least 2x if possible... I wonder if there's a datasheet available for KT133A north bridge.

It can still be bad caps.
You presumably haven't checked/changed the SMD caps about the size of a grain of rice.

If you really want to get to the root of it, you'll need at a minimum a multimeter and patience.

You can also try the hack I've fixed a surprising amount of dead hardware with: Dishwasher, no soap. Dry COMPLETELY before testing.
I've never had it make anything worse than it was. Several have improved. Several unchanged.

Dirt and random corrosion are often in places only the erosive power of the universal solvent can reach.

Mike_ wrote on 2026-08-13, 05:49:

I wonder if there's a datasheet available for KT133A north bridge.

I would say you need a "VIA 'xxx' North Bridge Programming Guide,"
the WPCREDIT utility (and its companion WPCRESET) by H. Oda with the corresponding PCF file for your bridge,
And a ton of patience.

I had a similar problem with a Abit KT7A some time ago, faster AGP cards were crashing/hanging the system a lot, especially in 4x mode. I don't remember how I solved this in the end, I think it was bad SDRAM or wrong memory timings.
If your BIOS has an option "Video RAM Cachable" try disabling that.

All of my Abit KT7A do not exhibit this problem. I tested with geforce 7600gt, an even later card that used a HSI bridge.

If you have a spare drive I think it might be worthwhile to install Windows XP or 2000 for diagnostic purposes. When I installed XP on my KT7A in 2002 it halted during the installation, and I discovered that my generic 256 meg stick of SDRAM was the issue because I had set it to high performance timings and it couldn't handle it. This issue had no doubt affected the stability of Windows 98, but I thought it was a Windows 98 problem rather than a self inflicted hardware issue.

myne wrote on 2026-08-13, 06:31:

It can still be bad caps.
You presumably haven't checked/changed the SMD caps about the size of a grain of rice.

If you really want to get to the root of it, you'll need at a minimum a multimeter and patience.

I doubt it's caused by ceramic capacitors. Usually they won't go bad, and in any case the Geforce 2 MX 400 is working fine at 4x AGP.

shevalier wrote on 2026-08-13, 06:35:

I would say you need a "VIA 'xxx' North Bridge Programming Guide,"
the WPCREDIT utility (and its companion WPCRESET) by H. Oda with the corresponding PCF file for your bridge,
And a ton of patience.

Unfortunately a programming guide doesn't seem to be available. 😁 Datasheet apparently is, and it at least somewhat explains what the hex value means. Upper part of value is apparently N side and lower P side of the buffer.

https://www.alldatasheet.com/datasheet-pdf/pd … C2/VT8363A.html

I tried WPCREDIT and apparently "AUTO" sets it to 0xDD on 1x and 2x mode and to 0xDC at 4x. This happens both with GF2 MX 400 and FX 5900XT.

Maybe it could be useful if somebody else checks what that register is set to on their motherboard?

Repo Man11 wrote on 2026-08-13, 14:38:

If you have a spare drive I think it might be worthwhile to install Windows XP or 2000 for diagnostic purposes. When I installed XP on my KT7A in 2002 it halted during the installation, and I discovered that my generic 256 meg stick of SDRAM was the issue because I had set it to high performance timings and it couldn't handle it. This issue had no doubt affected the stability of Windows 98, but I thought it was a Windows 98 problem rather than a self inflicted hardware issue.

I have tested the system with memtest86 and prime95 in blend mode. It sounds unlikely that memory would be quite *that* flaky and still pass those tests.

Power?
Mx =low watts
Fx =higher
Bx era =low agp slot wastage
Try the fx underclocked at 4x

Mike_ wrote on 2026-08-13, 14:47:

I doubt it's caused by ceramic capacitors. Usually they won't go bad, and in any case the Geforce 2 MX 400 is working fine at 4x […]
Show full quote

myne wrote on 2026-08-13, 06:31:

It can still be bad caps.
You presumably haven't checked/changed the SMD caps about the size of a grain of rice.

If you really want to get to the root of it, you'll need at a minimum a multimeter and patience.

I doubt it's caused by ceramic capacitors. Usually they won't go bad, and in any case the Geforce 2 MX 400 is working fine at 4x AGP.

shevalier wrote on 2026-08-13, 06:35:

I would say you need a "VIA 'xxx' North Bridge Programming Guide,"
the WPCREDIT utility (and its companion WPCRESET) by H. Oda with the corresponding PCF file for your bridge,
And a ton of patience.

Unfortunately a programming guide doesn't seem to be available. 😁 Datasheet apparently is, and it at least somewhat explains what the hex value means. Upper part of value is apparently N side and lower P side of the buffer.

https://www.alldatasheet.com/datasheet-pdf/pd … C2/VT8363A.html

The attachment agp_regs.gif is no longer available

I tried WPCREDIT and apparently "AUTO" sets it to 0xDD on 1x and 2x mode and to 0xDC at 4x. This happens both with GF2 MX 400 and FX 5900XT.

Maybe it could be useful if somebody else checks what that register is set to on their motherboard?

Repo Man11 wrote on 2026-08-13, 14:38:

If you have a spare drive I think it might be worthwhile to install Windows XP or 2000 for diagnostic purposes. When I installed XP on my KT7A in 2002 it halted during the installation, and I discovered that my generic 256 meg stick of SDRAM was the issue because I had set it to high performance timings and it couldn't handle it. This issue had no doubt affected the stability of Windows 98, but I thought it was a Windows 98 problem rather than a self inflicted hardware issue.

I have tested the system with memtest86 and prime95 in blend mode. It sounds unlikely that memory would be quite *that* flaky and still pass those tests.

I didn't mean to imply that you have a memory issue; my point was that I had a self inflicted hardware issue that was revealed by a different operating system.

Possibly a more relevant comparison: I bought a Quadro 4 700 that needed the capacitors replaced, recapped it, and tested it, but once you installed the driver you got the image below and the system locked up. I was pretty sure the card was junk, but as a Hail Mary I tried XP and it worked perfectly. That was a relief, but it was still a mystery why it wouldn't work with Win98. Then I discovered that it worked with Win98 in an MVP3 motherboard with AGP 2x. What I finally figured out was that it would work fine with AGP 4x and Windows XP/2000, and it worked fine with Win98 unless it was set to 4x. It was a Dell OEM card and I found a Dell BIOS update for it that allowed it to work with Win98 and AGP 4x. If I hadn't tried it out with XP I likely would have just given up on it and written it off.

Quadro 4 700 doesn't work with Windows 98 SE

myne wrote on 2026-08-13, 14:59:

Power? Mx =low watts Fx =higher Bx era =low agp slot wastage Try the fx underclocked at 4x […]
Show full quote

Power?
Mx =low watts
Fx =higher
Bx era =low agp slot wastage
Try the fx underclocked at 4x

Probably not. FX 5900XT has a molex connector for power, so it shouldn't draw that much from the motherboard.

Mike_ wrote on 2026-08-13, 14:47:

and it at least somewhat explains what the hex value means.

That's the exact datasheet I meant.
But it has a million registers and makes zero sense as to what's actually supposed to go in there. 🙁
That's why I wished you tons of patience. 😀
Well, or comparing it to another motherboard from a different manufacturer.

shevalier wrote on 2026-08-13, 16:51:

That's the exact datasheet I meant. But it has a million registers and makes zero sense as to what's actually supposed to go in […]
Show full quote

That's the exact datasheet I meant.
But it has a million registers and makes zero sense as to what's actually supposed to go in there. 🙁
That's why I wished you tons of patience. 😀
Well, or comparing it to another motherboard from a different manufacturer.

Yes, trying to decrypt unclear datasheets isn't a fun way to spend spare time. 😁

The latter sounds much better, so it would be really nice if somebody with a KT133A board functioning at 2x or 4x AGP with newer Geforces would dump the registers. 😀

Here's what you want.
https://www.overclockers.com/wpcredit-hacking … f-kt133-boards/

Someone already went over the datasheet and decoded it centuries ago

I'd still try underclocking it. It'll cost you 10 minutes and you can cross it off.

myne wrote on 2026-08-14, 03:01:

Here's what you want. https://www.overclockers.com/wpcredit-hacking … f-kt133-boards/ […]
Show full quote

Here's what you want.
https://www.overclockers.com/wpcredit-hacking … f-kt133-boards/

Someone already went over the datasheet and decoded it centuries ago

I'd still try underclocking it. It'll cost you 10 minutes and you can cross it off.

That page doesn't really tell anything useful regarding AGP.

I also tried different FX series cards that draw less power, such as the FX 5600 ultra that I had borrowed and a regular FX 5600. The same thing happens, so it can't be a fault of an individual card.

The chipset...