Sure does look like one of the two has bad memory if they are in SLI, if they are both doing it in single card mode then they both have memory IC problems. My guess is there is a bad memory IC on one of the two TMUs as Frame buffer memory would give a different type of corruption.

They each display this problem in single card mode, although with a slightly different pattern.

I'll have to find some replacement memory that is compatible.

Do you have any other images of the issue? If it's just colouring, I'd lean more towards the analog circuit (eg. capacitors)

I've already disconnected and packed it up. The dot pattern also exists on textures in Quake 1, which makes me feel it's something wrong with the texture memory, or maybe a loose pin on a TMU.

Kahenraz wrote on 2024-02-24, 08:45:

I've already disconnected and packed it up. The dot pattern also exists on textures in Quake 1, which makes me feel it's something wrong with the texture memory, or maybe a loose pin on a TMU.

Its weird they both got the same issue at the same time .. is it possible they got a bit more angry pixies than they should have ? Even a slightly higher PCI voltage than spec may cause older ICs to fail.

Though if both cards were fabbed around the same time old tired solder might be worth checking out.

I believe that both of these were fully dead at some point and I revived them by reflowing all of the large QFPs. I seem to recall the legs having cracked and being manipulatable with a probe. One of them has bent legs that I also repaired.

These were one of my first large QFP repairs, and I've gotten a lot better since then, so maybe they need to be revisited. I cleaned up the legs and added fresh solder though, so it would be odd if it's the same problem as before. They have since been living in a small bin within their own individual antistatic bag and were well protected.

Edit:

I found the repair log here:

Voodoo 2 triage and repair

This thread is actually from one of my working cards, but the dot pattern can be seen in one of the photos where I had a solder bridge on a TMU. This suggests that it might be loose QFP legs, like I thought. I feel that these components work themselves loose if time due to repeated thermal expansion, since they have no heatsink and tend to run very hot.