i have an MSI K8MM-V that has this bug, oldest bios version dates 2005 so i can't downgrade to a version that doesn't have the bug
the chipset is a via K8M800, installing windows XP doesn't fix it on this board either

I have a K8V SE non Deluxe that has the bug and I can't find a suitable Bios or Afudos that will let me flash backwards.

I've just checked one of my via boards for this, its an MSI PM8PM-V, MS-7222 Ver2.0, no problem on bios version 3.3, just updated to latest bios Ver 5.5, again all ok no bug.

Hi,
can't confirm the bug on an Asus K8V MX/S board with a VIA K8M800 chipset.

My setup:
Athlon 64 3400+ Clawhammer
Radeon 9500 64 MB with a universal AGP key - Catalyst 6.2
2 GB DDR SDRAM (Thank you, Mr Loew!)
BIOS version: 08.00.09, revision 1008.010, build date 05/12/05
Windows 98 SE

The system scored over 11 000 3DMarks in 3DMark2001SE, which is AFAIK acceptable. The AGP memory size is correctly displayed in 3DMark's information box.
Cheers!

ShyShutterbug wrote on 2024-04-22, 21:21:

Hi, can't confirm the bug on an Asus K8V MX/S board with a VIA K8M800 chipset. […]
Show full quote

Hi,
can't confirm the bug on an Asus K8V MX/S board with a VIA K8M800 chipset.

My setup:
Athlon 64 3400+ Clawhammer
Radeon 9500 64 MB with a universal AGP key - Catalyst 6.2
2 GB DDR SDRAM (Thank you, Mr Loew!)
BIOS version: 08.00.09, revision 1008.010, build date 05/12/05
Windows 98 SE

The system scored over 11 000 3DMarks in 3DMark2001SE, which is AFAIK acceptable. The AGP memory size is correctly displayed in 3DMark's information box.
Cheers!

Which chipset driver version are you using? I think that last time I tested, 1008 was giving me the AGP bug on mine...

Regards

Chipset driver: VIA Hyperion Pro V521A
ACPI is on, FDD, serial and parallel ports disabled (I've disabled those to free up some IRQs for a Yamaha YMF724 soundcard)
All USB ports enabled, USB 2.0 driver version: VIAUSB2V270-L-M

I was actually quite surprised to see it work, considering many dramatic reports here 🤣 Since you got different results, I'll see if I can reproduce the bug by any simple tweaks.

Cheers!

My understanding is that this 0 bytes texture memory is more of a Via + Nvidia bug than it is only VIA's bug.
From reports online and my personal experience it doesn't happen on ATi cards.

At least on one of my systems with Asus M2V which is PCI-e system I don't get texture memory on Windows Me with Quadro FX1300 and performance is much worse than on XP/Win7 32-bit (with XP's drivers). Not even things which you would think would be affected - which is more like a Voodoo/Vooodoo2 behavior so fine until you hit memory limit and/or at least you could use lots of AA/AF. Performance is just much worse overall. Its below 10K points... probably still enough for games which need Windows 9x and FX's features but still.

Anyways on Radeon X800XL 256MB I don't get this issue and 3dMark2001SE scores almost the same as on XP. There is like 1GB texture memory reported. Of course X800XL is much faster GPU anyways and in some sense even better because its image quality is better. I guess that if I really need to use this computer and want to play old DX/OGL game that doesn't run in Windows XP/7 and needs FX and runs fine on its drivers then it should still have enough performance...

What I tried is different VIA 4in1 versions. Different BIOS revisions down to 0302 which required putting Sempron 3200+. I played with PCI texture memory in Riva Tuner and then I broke my fully configured Windows ME with DOS mode and 2GB memory support by pointlessly changing some settings in "PCI bus" device for controlling PCI resource allocation. Safe mode doesn't exactly work so I will need to reinstall system or figure out how to fix it. Of course nothing helped.

What I didn't try but someone with ability to recover (read: flash using external programmer) + motherboard where BIOS downgrade fixes the issue might try is to create custom BIOS hybrids to identify which component of the BIOS is responsible for the bug. Chances are it would be possible to at least have solution for boards which already don't exhibit the bug with older BIOSes but also be able to use newer AGESA and thus newer CPU's not being limited to the early 2005 CPU's. It might also be possible that module responsible for this issue is universal enough than even newer boards which don't have version without the bug to be able to have full performance on Nvidia cards.

BTW. From success (as its VERY IMPORTANT to share those!!) stories I also have Gigabyte K8VT800 rev.2.0 which is socket 754 mobo and I have newest BIOS "FG" installed with support for Venice CPU's. I have Win98SE installed there and observed no issues with GeForce FX5900ZT and GeForce 4Ti 4200 AGP8x.

Deleted because it was useful but ignored.

On the ASUS K8v-MX, I narrowed down the specific BIOS that introduced the bug, 0203. I've flipped in between 0112 and 0203 and the bug follows 0203. Since VIA's AGP driver has debugging symbols, I'm attempting to learn how to setup SoftICE and trace out exactly where the GART is failing.

Deleted because it was useful but ignored.

myne wrote on 2024-12-02, 01:28:

Were there any differences in the registers?
Did you manage to find datasheets for the chipset(s)?

In the AGP specific registers (D0F0, D1F0), no, they are identical. I neglected to check the GPU's registers though. I'll report back with those results.

I couldn't find a datasheet for this chipset, but Intel's PCI/AGP standard documentation provides a lot of useful information.

Deleted because it was useful but ignored.

Deleted because it was useful but ignored.

Deleted because it was useful but ignored.

Any progress?

myne wrote on 2024-12-02, 02:12:

Dump every register for everything.
For all we know it could be a pci setting.

The key is to spot any and all differences between the versions.

I dumped every register, except USB, RAID, and IDE devices, and there are only 2 anomalies: device 24 function 2 and function 3. Screenshots below of their differences.

In function 2, the bytes at 0xB0-0xB4 change every time I open WPCRE. I don't think this is relevant.

In function 3, the byte at 0x9C can switch between 02 and 00 whenever I reopen WPCRE, regardless of BIOS version. Same story at byte 0xE6.

I don't think the answer is in the registers. Even the GPU's registers are the same.

(Please don't spam threads with replies, just edit your first reply. PM me if you want more updates or other specifics.)

Deleted because it was useful but ignored.

myne wrote on 2024-12-04, 05:50:

They're both amd devices. […]
Show full quote

They're both amd devices.

You need 1106 for via.

https://devicehunt.com/view/type/pci/vendor/1022/device/1102

https://devicehunt.com/view/type/pci/vendor/1022/device/1103

This should be it

https://devicehunt.com/view/type/pci/vendor/1106/device/2204

Like I said, I dumped the registers of every device WPCRE would let me, that weren't some kind of storage or USB controller. 1106-0204, 1106-1204, 1106-2204, 1106-3204, 1106-3227, 1106-4204, and 1106-7204 were all the VIA devices listed. 10DE-0172 is the Nvidia MX 420, again no differences between BIOS versions.

.Deleted because it was useful but ignored.