Card1(Quantum3D 100SB) :

-removed the Shade modules to see if it makes any difference, it didn't.
-the card is correctly detected by mojo.exe and by the q3d utility:

-inspected the card under the microscope, everything looked fine. Checked all TMU and FBI pins and all were OK. Measured all the resistors on the board and all were OK as well.
-started the card up and checked temperatures, all good:

-disabled the TMUs from autoexec.bat by adding "set SST_TEXMAP_DISABLE=1" (it can be any value, doesn't matter), checked again, still running a 3D application causes a freeze.
-disabled SLI from autoexec.bat by adding "set SST_SLIDETECT=0", to narrow the issue down a bit, still not working.

This means that the issue is located on this part of the card (circled in red) and most likely FBI and FBI RAM related (circled in yellow):

These cards use two types of RAM, FBI has 512KB ICs, the TMUs have 1MB ICs.

Since I already checked all the FBI pins and they were OK, I checked all the RAM pins next. This pin was not connected to the pad underneath it:

This was the RAM IC with the unsoldered pin:

After soldering it, now the card works:

Since now I have a working card, I tested the 4 Shade modules on it, and all work fine:

Quake2 demo1.dm2 timedemo, 640x480, with the extra TMUs: 44.1FPS, without the extra TMUs: 32.7FPS.

Card2(Quantum3D 100SB) :

It either prevents Windows from starting, or, when it does start, not even the PCI to PCI bridge IC is detected. Under DOS, running mojo, no cards are detected.

This can be caused by a number of things:
-broken traces from the PCI connector to the PCI-PCI bridge IC
-broken PCI-PCI bridge IC
-broken FBIs (some PCI-PCI or PCIe-PCI bridge ICs won't even be detected by the OS, if whatever device is after them screws with the PCI bus hard enough)
-PCI-PCI bridge IC - missing power rails

I first checked various PCI traces and components near the bridge IC, all looked and measured fine. Checked the bridge IC pins (will come back to that...), all were soldered properly.
While most ICs on this board are powered by 5V, the 21152 PCI-PCI bridge needs a 3.3V rail, which is provided by this IC:

Measured this and it did provide 3.3V to the bridge IC. At this point, it's either that IC or the 3dfx FBIs.

I decided to try swapping the bridge IC first. I don't have any spares, but luckily I got this Matrox card:

This one is Intel branded, and it's an 21152-AB instead of 21152-AA, but it should work fine. If everything works, I'll try to source some "digital" branded ones and replace it in the future.
Since I had to remove both ICs, my initial plan was to swap them first, add the one from the 100SB on the Matrox card, and the one from the Matrox card on the 100SB.
This is how the IC on the 100SB looked before I removed it:

And this is how it looked after I removed it:

There were also broken pins on opposite corner of the IC. This wasn't caused by me removing the IC, the pins were already broken.

After replacing the bridge IC, I started the system and ran mojo.exe:

All looks good, and it's also properly detected by the Q3D utility.

Ran Quake2, and, the game freezes....

As I did on card #1, I disabled SLI and the TMUs and tested again, the application would still freeze. I then checked resistors, FBI pins, traces, and all looked fine.
Started measuring FBI RAM again, and guess what? Same unsoldered pin on the same RAM IC as on the #1 card:

This is extremely weird, but I'll take it. Soldered it back and:

So the first FBI seems to work properly now. Enabled the TMUs but not SLI:

If I enable SLI, with or without the TMUs enabled, the whole system freezes.

So there is another fault, but at least this part of the card now works:

Nothing else was wrong with the card, when I reenabled SLI, it just happened to freeze because of the S3 card that seems to be on the way out.

EDIT: to clarify this a bit. Desktop resolution was set to 1024x768, and Q2 was set to run in accelerated mode. When Q2 is started, even if it is rendered by the 100SB card, when starting the game the Desktop resolution (from the S3 card) is changed to 640x480, and reverts to 1024x768 when the game is closed.. This is how Q2 behaves, and this change in 2D resolution caused the system to freeze.

Here it is with SLI enabled and the Shade modules installed:

Those cards look amazing, well done on the fixes 😀 Dual slot-1 CPUs and that quantum card look fantastic in thermal vision too, surely that gets pretty toasty after not long.

That damage to the PCI bridge took me a moment to figure out what it was that was wrong but that's due to the microscope perspective. I wonder what kind of movement or impact caused that damage?

Thanks! They do look amazing 😁
The system gets a bit toasty if the side panel isn't installed, otherwise it's fine, due to the 3 fans installed on the front of the case.

Regarding the PCI bridge IC damage, it's really weird. It doesn't look like it was ever moved or impacted, and impact hard enough to break those pins would have twisted all the other pins, but everything looked fine. I inspected the IC under the microscope, and there was no sign of any kind of damage.
It looks like metal fatigue, but I have no idea how that could have happened.

Card3(Quantum3D 100SB) :

While initially this was the only somewhat working card, it took the longest to fix.

Output looked like this:

As on the other cards, I disabled SLI. The issue was still present, although it looked a little better. The major pain repairing this was that the problem was intermittent.
Everything would look fine for hours and then looked similar to the above picture, or looked similar to the above picture and then started working fine.

Tried making the card going from one state to the other, applied pressure on ICs, flexed the card, tried, one at a time, cooling or heating every component. I could never trigger a change.
Another interesting thing, when going from not working properly to working properly, it would only do so when a new Quake 2 level was loaded.

Inspected the card under the microscope, everything looked OK. Measured all resistors and inductors on the card, all looked good. I only found 1 broken FBI decoupling cap, which wouldn't cause this kind of errors.
Since the issue was intermittent, it wouldn't make much sense checking all RAM pins for continuity, so I redid almost all solder joints in this area of the card:

No change.

At this point I started suspecting the TMU and the RAM ICs. I decided to replace the TMU at this point.
After it was replaced, I tested the card extensively, and it worked fine (still without SLI enabled). Just to be sure that the original TMU was the problem, I removed the replacement and soldered back the original TMU.
With the original TMU in place, it starting behaving exactly as before.

Soldered the replacement TMU back on, tested the cards, worked fine again. Enabled SLI, and it still had some graphical issues, although it looked way better than before.

Next I just redid all second TMU RAM solder joints, and now the card works fine.

And with Shade modules installed:

I ran a couple of Q2 benchmarks with and without extra TMUs, with and without SLI enabled:

No SLI:
(1x FBI + 1x TMU) x1 21.4FPS
(1x FBI + 2x TMU) x1 28.7FPS

SLI:
(1x FBI + 1x TMU) x2 32.7FPS
(1x FBI + 2x TMU) x2 43.8FPS

Excellent work!

Would also love to get my hands on one of those! 😀

Wow, fantastic work! It's interesting to see how how that extra TMU helps almost as much SLI. Multitexturing was a big deal

@Paadam @DrAnthony Thanks!

The DEC branded PCI-PCI bridge ICs arrived today:

I replaced the Intel branded one on the #2 card:

And a last group photo, now all working properly:

Nice one! And even matching the date code (mid-1997) 😀

Where the pins on the digital chip broke, they look thinner than on surrounding ICs eg the GENDAC, and happen to be near the PCI connector. I wonder whether it ould simply be an area of high physical stress when the card is being inserted (especially if it gets slightly flexed) and/or could that also happen when it heats up - with the two ends of the card and the PCI slot being held by the case.

@havli Yes, the datecodes were a nice surprise 😀

@ratfink It is possible. To me it looks like metal fatigue, the IC wasn't bumped/twisted/hit etc. It looked pristine. Weird fault really, I haven't ever seen something like this.

Card4(Quantum3D Obsidian2 X-24):

This was previously posted here: Quantum3D Obsidian2 X-24 repair

The card worked fine without the daughterboard installed, or with a know good daughterboard. At a first look, it didn't look that bad, there were a few scratches but no broken traces.

However, the right TMU, TMU0, showed signs of corrosion on about 25% of the pins.

The legs/pads would not take new solder, regardless of the amount of flux used, so I removed it to see what the actual damage is.

While the pads are still there, more than 90% of the affected ones don't have a connection to the traces anymore.

I proceeded to scrape off the soldermask and the corrosion until I could see bare copper. And it was worse than I expected.

Next I checked if the vias were still connected to the corresponding TMU1 and FBI pins, as well as GND and VCC. Most of them were salvageable.

The pads were really corroded even underneath, and when touched by the soldering iron, they would instantly lift. So, wire and liquid soldermask it is. Solder a couple, use soldermask, UV cure, and repeat.

After all pads were fixed, and the soldermask cure, I scraped off the soldermask where the new TMU would be soldered:

Soldering the new TMU was a bit of a pain, but doable:

There was also a broken 10R resistor array, and after fixing that the card runs nicely in SLI.

Card5(Quantum3D Obsidian2 200SB):

Got this card a while back, sold as working. Sometimes it works, but most of the times it either prevents the system from starting up, or does this:

Taking a look at the card, someone replaced the PCI-PCI bridge IC, and did a rather poor job of it. Two pads were ripped, a bunch of capacitor are simply missing, and a few are crooked:

I removed the bridge IC (which had a lot of gunk under it), replaced the missing caps, and proceeded to fix the missing pads (they were not connected to anything, this is just for aesthetic reasons).