Could be a resource conflict with the PEX chip. Might be "fixable" by trying out a bunch of different AGP boards.

From the screenshot, it seems like something corrupted on the surface (pci.sys or related stuffs).

If the driver files were actually intact, maybe try fiddling the "AGP Aperture Size" settings. It controls the size of AGP GART which might be the cause of memory-related conflicts, especially with older AGP 2x (3.3V) boards.

I analyzed this sandwich, figured out the topic, and AI also gave me a lot of ideas. I think that this design is not workable, and I will explain why:
-The PCI to PCIe adapter operates at an extremely high frequency of 66 MHz, so data between the MB and the adapter is likely to be corrupted, which is why pci.sys was damaged, so it's not a driver issue.
- The AGP adapter is passive without a bridge. If there was a bridge that could lower the frequency for the PCI to PCIe adapter, this sandwich would work, but it would be an exchange of data between three bridges: north bridge to bridge (AGP to PCI) to bridge (PCI to PCIe), which would cause a large delay and leave little of the theoretical 266 MB/s, possibly less than 150 MB/s of the PCI bus.

Conclusion: adequate operation of this sandwich is unlikely.

AGP aperture should not matter when operating it as a fast pci bus.

The file contents of pci.sys is not touched by the agp bus.

st31276a wrote on 2026-03-06, 09:49:

AGP aperture should not matter when operating it as a fast pci bus.

The file contents of pci.sys is not touched by the agp bus.

Yes, that's right, but the file was corrupted, likely just from trying to read it, because WinXP was taken from one of my retro builds, and before the experiment, I loaded the OS with a regular card on this MB.

After all, my conclusions are incorrect. I checked the bridge datasheet, and it actually operates at a frequency of 66 MHz. In this case, it is unclear why the system files were damaged.
https://www.alldatasheetru.com/datasheet-pdf/ … 12-AA66BIF.html