It will not damage the card. Will it fix the problem? Thats a different question.

1.use old MATS/MODS for dos to get which memory module does not work.
https://drive.google.com/drive/folders/1pFuLL … kJDv4TiJKE_GWNw

2. i have a GF3 where is 4 modules (3.6ns) died on me when i did the 3dmark tests.
i replaced with 4ns modules from another GF3 and everything works just fine.
so i have 4 modules 4ns and 4 modules 3.6 on the same board.

All else being equal, if they're just slower, you might need to downclock the memory.
I'd stick to the same brand though, just to minimise the chances of other issues.

dm- wrote on 2024-07-05, 04:04:

1.use old MATS/MODS for dos to get which memory module does not work. https://drive.google.com/drive/folders/1pFuLL … kJDv4TiJKE […]
Show full quote

1.use old MATS/MODS for dos to get which memory module does not work.
https://drive.google.com/drive/folders/1pFuLL … kJDv4TiJKE_GWNw

2. i have a GF3 where is 4 modules (3.6ns) died on me when i did the 3dmark tests.
i replaced with 4ns modules from another GF3 and everything works just fine.
so i have 4 modules 4ns and 4 modules 3.6 on the same board.

Golden! Didn't know such old MODS/MATS ever leaked! Now I need to source a floppy. I suppose I need to mount the MATS image and run it. What order are the chips with respect to their physical position?

I think that you can run the video card without all memory chips connected since I ran my broken GF4 Ti4200 with two memory chips removed and it just has some thick white bars where those chips would have held framebuffer.
But possibly some later cards like the FX series don't tolerate that in the same way

Jofo wrote on 2024-07-09, 19:53:

dm- wrote on 2024-07-05, 04:04:

1.use old MATS/MODS for dos to get which memory module does not work. https://drive.google.com/drive/folders/1pFuLL … kJDv4TiJKE […]
Show full quote

1.use old MATS/MODS for dos to get which memory module does not work.
https://drive.google.com/drive/folders/1pFuLL … kJDv4TiJKE_GWNw

2. i have a GF3 where is 4 modules (3.6ns) died on me when i did the 3dmark tests.
i replaced with 4ns modules from another GF3 and everything works just fine.
so i have 4 modules 4ns and 4 modules 3.6 on the same board.

Golden! Didn't know such old MODS/MATS ever leaked! Now I need to source a floppy. I suppose I need to mount the MATS image and run it. What order are the chips with respect to their physical position?

I've tried it out and it works for all cards from the Geforce 1 up to the FX series, but not NV4x / 6000 series and later. It's a bootable DOS floppy which autoexecs the MATS test program.

If someone knows which chips correspond to which channel for common cards, I'm also interested. You could try putting a fault on known memory chips to find out the layout which is fairly easy with TSOP chips since you can put a blob of solder connecting a data pin to a neighbouring ground pin, then log which channel / bit area i.e. FBA 31:0 that the fault was detected as - I mapped out the memory channels on the Voodoo 3 here with this method: Re: Accurately troubleshooting video memory faults with VMTCE
MATS is nice because its output is even more detailed than what VMTCE can achieve, here's some examples of tests on a bunch of my cards

Do you guys know which version of mats works on GTX 465/470? mats version 367.38 works fine on GTX 480 but it shows weird memory bank letters for GTX 465/470:

GTX 465 starts with Bank C:

1  SUBPART     RANK0 RD ERR  RANK0 WR ERR   UNKNOWN ERR
2------------- ------------- -------------  ------------
3FBIOC[ 31:  0]            0             0             0
4FBIOC[ 63: 32]            0             0             0
5FBIOD[ 31:  0]            0             0             0
6FBIOD[ 63: 32]            0             0             0
7FBIOE[ 31:  0]            0       1356877             0
8FBIOE[ 63: 32]            0             0             0
9FBIOX[ 31:  0]            0             0             0
10FBIOX[ 63: 32]            0             0             0

GTX 470 starts with Bank B:

1FBIOB[ 31:  0]            0             0             0
2FBIOB[ 63: 32]            0             0             0
3FBIOC[ 31:  0]            0             0             0
4FBIOC[ 63: 32]            0             0             0
5FBIOD[ 31:  0]            0             0             0
6FBIOD[ 63: 32]            0             0             0
7FBIOF[ 31:  0]            0             0             0
8FBIOF[ 63: 32]            0          8080             0
9FBIOX[ 31:  0]            0             0             0
10FBIOX[ 63: 32]            0             0             0

GTX 480:

1FBIOA[ 31:  0]            0             0             0
2FBIOA[ 63: 32]            0             0             0
3FBIOB[ 31:  0]            0             0             0
4FBIOB[ 63: 32]            0        879264             0
5FBIOC[ 31:  0]            0             0             0
6FBIOC[ 63: 32]            0             0             0
7FBIOD[ 31:  0]            0             0             0
8FBIOD[ 63: 32]            0             0             0
9FBIOE[ 31:  0]            0             0             0
10FBIOE[ 63: 32]            0             0             0
11FBIOF[ 31:  0]            0             0             0
12FBIOF[ 63: 32]            0             0             0

Try 343.96: https://drive.google.com/file/d/17wOPsv60-9AS … iew?usp=sharing

Masterchief79 wrote on 2024-07-10, 19:17:

Try 343.96: https://drive.google.com/file/d/17wOPsv60-9AS … iew?usp=sharing

Thanks, this seems to work.
GTX 465:

1    LANE     EB0 READ ERR EB0 WRT ERR  UNKNOWN ERR 
2------------ ------------ ------------ ------------
3FBIOB[31: 0]           0            0            0
4FBIOB[63:32]           0            0            0
5FBIOC[31: 0]           0            0            0
6FBIOC[63:32]           0            0            0
7FBIOD[31: 0]           0            0            0
8FBIOD[63:32]           0            0            0
9FBIOE[31: 0]           0      1356880            0
10FBIOE[63:32]           0            0            0

GTX 470:

1FBIOB[31: 0]           0            0            0
2FBIOB[63:32]           0            0            0
3FBIOC[31: 0]           0            0            0
4FBIOC[63:32]           0            0            0
5FBIOD[31: 0]           0            0            0
6FBIOD[63:32]           0            0            0
7FBIOE[31: 0]           0            0            0
8FBIOE[63:32]           0            0            0
9FBIOF[31: 0]           0            0            0
10FBIOF[63:32]           0         8080            0

Yeah nice. Hopefully a memory reflow is enough to make them work again.
I posted a quick video on that once just to show how I'm doing it: https://www.youtube.com/watch?v=no1W3bJFnVs

Also big thanks to the legend who posted the old Mats/Mods versions. Now I can finally diagnose GeForce4 cards and older. One 4200Ti is already working again and I have like 5 more cards here waiting to be repaired. Thanks again!

Might be worth uploading to the driver site

Good hint with mixing modules btw, I have this card in for repair and could revive it by stealing two 4.0ns EliteMT Chips from Medion GF3.

dm- wrote on 2024-07-05, 04:04:

1.use old MATS/MODS for dos to get which memory module does not work. https://drive.google.com/drive/folders/1pFuLL … kJDv4TiJKE […]
Show full quote

1.use old MATS/MODS for dos to get which memory module does not work.
https://drive.google.com/drive/folders/1pFuLL … kJDv4TiJKE_GWNw

2. i have a GF3 where is 4 modules (3.6ns) died on me when i did the 3dmark tests.
i replaced with 4ns modules from another GF3 and everything works just fine.
so i have 4 modules 4ns and 4 modules 3.6 on the same board.

Awesome, I didn't know such an old MATS version existed in the wild. I've managed to quickly fix a GeForce 4 Ti 4200 card using it, so thank you. 😀

FYI: I initially got an "invalid opcode" error during the MATS startup. Turns out, it was an incompatibility with the VIA C3 Ezra-T CPU I was using on the 440BX motherboard I was testing with. After replacing the CPU with a Pentium 3 Katmai, MATS worked like a charm.

Finally was able to get some floppies and decided to play around with it. However, the result is a bit surprising, seems like almost all modules are bad

A1, B0, C0, C1, D0, D1 seem bad, leaving just A0 and B1 okay. Besides, I don't even know which is which, so will probably just swap all of them

Jofo wrote on 2024-09-10, 20:46:

Finally was able to get some floppies and decided to play around with it. However, the result is a bit surprising, seems like almost all modules are bad

The attachment GF3.jpg is no longer available

A1, B0, C0, C1, D0, D1 seem bad, leaving just A0 and B1 okay. Besides, I don't even know which is which, so will probably just swap all of them

Look for broken components like resistors around the card. If that doesn't help then you can check your bad data lines against the schematic: https://www.phantom.sannata.org/download/file.php?id=10440

The bad data lines in your screenshot seem to be all over the place so it could be one missing capacitor somewhere fouling up all the memory data lines. If it was just one or two data lines at specific locations you could figure out if it's a bad chip or a bad pin on the BGA.
Perhaps you have some bad / damaged resistors for some of the memory chips like for instance if the vref on the DDR memory is broken then the DDR memory can't operate.

Worth attempting to underclock it a lot and see if it changes.
If it passes at a very low speed, then at least everything is contiguous.

Errors on all channels can also mean a dead memory controller (inside of the GPU). It's pretty unlikely that all memory chips are bad, if the card didn't have previous damage in the memory VRM or something along those lines.
Look very closely for damage on the PCB, check if it's straight, swap one memory channel if you don't find anything, and test the card again. If the errors don't change, the card needs either a GPU reball or a new GPU altogether.

Masterchief79 wrote on 2024-09-11, 07:47:

Errors on all channels can also mean a dead memory controller (inside of the GPU). It's pretty unlikely that all memory chips are bad, if the card didn't have previous damage in the memory VRM or something along those lines.
Look very closely for damage on the PCB, check if it's straight, swap one memory channel if you don't find anything, and test the card again. If the errors don't change, the card needs either a GPU reball or a new GPU altogether.

I think it's a broken component issue, just found this result in my MATS tests:

This is a 128-bit FX5200 that I picked up specifically because it is the card from the NV34 card schematics (P162):

It turned out that those memory errors were caused by a couple of broken off capacitors and resistors on the back of the card. I have another copy of this card but broken so I took them off of that and now it's 100% working.
These things usually weren't stored well and those little 0402 components can break off quite easily.