Hey there,

I just checked: The PCB is the same as the MS-8872, design references match.

I checked the schematic and looked at the CS (Chip select pins), they're either 0 or 1, I would assume based on the banks.

If that's the case, this is your mapping:

Hi,

Thank you for these precious informations.
So the faulty chip seems to be U803...

Cheers,
JB

Appears like it is, yes.

In case you perform a repair attempt, can you share the results here? It would be cool to have confirmation regarding the memory mapping.

Also, if it doesn't give you too much of a headache, I'd appreciate that you post your repair here as well. It would make a nice addition and those GeForce 4Ti are just as prone to the failing BGA memory as the ATI 9X00 Series is (no wonder, it's the same memory)

Thanks!

you have faulty memory lines going one by one 24-31.
check the resistors around memory chip in D row, both front and back sides. (counterclockwise)

if one of resistors damaged (use microscope, use tester) it may lead to this situation, 8 lines in row

Yep. 8 exact following bits look very suspicious.

Apparently, unlike their ATI 9X00 counter parts, the Geforce 4Ti do not have series termination resistors on the board. Instead you'll only find parallel termination resistors (against VTT). Interesting!

If you want to take a look at the parts that dm- mentioned, I've attached part of the schematic and their respective position on the PCB.

Cheers

Thank you so much for your help!
I will share my findings.
Do you know if there is a reference design schematic for this board?

bimole wrote on 2025-06-18, 20:43:

Thank you so much for your help!
I will share my findings.
Do you know if there is a reference design schematic for this board?

You can get the complete reference schematic right here: https://phantom.sannata.org/download/file.php?id=28626 😀

The fact that it's 8 bits of a 32-bit single memory chip made me think is it a BGA issue - it may well be! Check this out, there are two pads at the edge that gate i/o for DQ25 to DQ31:

If a BGA gets damaged it's usually going to be at an edge or a corner, these two pins being right at the corner makes me think that either one of these pads isn't connected and that's why the RAM isn't working. Reflowing or reballing the RAM chip in question may resolve this.
That is assuming of course that the DQ0-31 numbers aren't all scrambled by the board layout and BGA fanout, I'm hoping that the MATS software translates that back into meaningful data line locations.

Edit: looking at the schematic, FBDD25 to 31 are linked to DQ0 to DQ7 of U803, perhaps the bad BGA balls are at the other corner?

They're also at the corner so still pretty likely it's a bad DQS (data strobe) or DM (data mask) connection that's breaking the RAM functionality.

In any case, have you checked if applying pressure to the memory chip changes behavior?

Cheers

chip reflow may help in 90% cases

There is a software tool that can test, and rewrite some older nvidia bioses to drop ram channels.
Eg it can make a 128bit card into a 96 bit card at the obvious cost of capacity.
I can't remember the name, or how far back it goes, but I believe it could do that bad bga era ~06ish and probably others.

Hi,
Some quick news...
I managed to install the drivers (45.23 on XP) despite strong artifacting.
And yes, if I squeeze U803 and PCB between my fingers, artifacts vanish. I even could run 3dmark2000.
So it's definitely U803. I will try to reflow it with some flux.

To be continued 😉

Cheers,
JB

hot air station, 350C. full airflow.
around 30-40 sec and chip will "float", gentle touch the chip by stick. done.

Lovely to hear. Good luck on the reflow! Glad to hear that you were able to nail it down!

Good idea to preheat it.

Even if you don't have a proper preheater, a normal oven ~100c 10-15 mins in your casserole dish should be good.

It reduces the stress of heat expansion which can be significant.

Don't burn yourself.

Hi guys,

It worked!!
Flux all around U803 chip and gently heated it with the air gun.
It's an air gun from an old Weller soldering station with a small air nozzle (a bit small compared to more recent air guns).
Despite I pushed it to the max temp (550C), I was still unable to gently move the chip, it was stuck. So the effective chip temp was not so high maybe...
Anyway, it work and I'm glad. Next time, I will try with my ATTEN ST-862D...

Will be soon because I have a second artifacting Quadro 980XGL!!
More to come with MATS reports...

Stay tuned 😉

Cheers,
JB

Cranking the heat that high is not recommended.

myne wrote on 2025-07-01, 22:34:

Cranking the heat that high is not recommended.

I agree. Often the PCB sinks a lot of heat away through the BGA balls, thus preheating helps a lot.

Yeah, it brings the delta down by at least 80*c

I know that 550C temp setpoint was quite high, but as I said, the nozzle was quite small. I didn't stay at the same place on the chip, always making circles above the area... No weird fumes or burnt flux residues so I think on average the temp was ok.
Preheat is a good idea... But I don't have any PCB heater.
I thought about a pan filled with sand with aluminum foil on top. Heated on induction cooker, the PCB laying on the aluminum foil, I think I could reach about 100C with a good time constant.
What do you think?