I realised that it's 16 of 32 bits.
So it's half a chip.
Is there a common signal to 1/2 a chip?

Not sure if I'm on to something but if I convert the errors to binary there's a pattern of 4 column bits.

Scratch that. Didn't notice the col was 3 characters so the first was overlapping.
It's 2 bits in Col Binary2.

While I'm here, here's the datasheet for the ram.
https://www.alldatasheet.com/datasheet-pdf/do … 323QF-GC20.html

And here's the name of a schematic file for a different 6800
nvidia_41-p260_nv41_256bit_ddr1_sch.pdf

If any experts drop by...

Seems like these boards are near enough identical.
https://www.techpowerup.com/vgabios/?did=10DE-0092-10DE-0301

More searching, finally found some things out.

MSI ran out of numbers with the last of the MS-89xx schematic numbering being these:
MS-8993-010 MSI Nvidia GeForce 6200TC 128MB
MS-8994 NX6600GT-VT2D128E Diamond
MS-8995-010. Radeon X800. 256MB,DDR3 SDRAM
?
MS-8997 Ver: 1 Radeon X800 Pro
MS-8998 FX5200-TD128LF
MS-8999 ATI Radeon 9250 128MB

Apparently they started again from V000.
And it seems like it is probably V033.

Which has several possible sources but all paywalled so far.
eg "MSI MS-V033-10 NV41_42_G70-P317, 256bit, 4M_8Mx32bit DDR3 boardview.7z"
"V033-10.cad Boardview (Bitmap)"
"Schematic VGA MS-Vxxx"

I've also looked around for a possible bit-mask in the bios. There is a tool for later cards https://gpuzelenograd.github.io/NVIDIA to disable memory channels, but not this far back, it seems so far.

Gotcha, bitch.
Anyone looking for NV41, NV42, or G70 based schematics look at this one. It appears to be the nvidia reference design P317/Q317. MS-V033
Hello to anyone googling from the long distant future.

All memory lines appear to be direct core to chip with no termination or pullups.

Also figured out that every single bad address has a 1 in bits 17,19,20,27

supposed to be chip no 2 counting from agp slot.
ddr3 chips have very weak pads, they may be ripped off just by desoldering from the board.
make sure you have spare parts.

What makes you think the 3rd?

sorry, second chip

I'm thinking it's the first.
Because Nvidia put their chips down anticlockwise with the A1, A0, B1, B0, C1, C0 order.
Which implies that the first chip is bits 32-63, and the second is 0-31, then B is 96-127, 64-95 etc.
The schematic and the board seem to agree that U14 (chip1) is the candidate.

Unless I'm reading it all wrong somehow.

myne wrote on 2026-05-13, 06:30:

I'm thinking it's the first. Because Nvidia put their chips down anticlockwise with the A1, A0, B1, B0, C1, C0 order. Which imp […]
Show full quote

I'm thinking it's the first.
Because Nvidia put their chips down anticlockwise with the A1, A0, B1, B0, C1, C0 order.
Which implies that the first chip is bits 32-63, and the second is 0-31, then B is 96-127, 64-95 etc.
The schematic and the board seem to agree that U14 (chip1) is the candidate.

Unless I'm reading it all wrong somehow.

From the schematic I agree:

1FBA: U13 (0-31) U14 (32-63)
2FBB: U11 (0-31) U12 (32-63)
3FBC: U8 (0-31) U10 (32-63)
4FBD: U2 (0-31) U4 (32-63)

No idea where it is positioned, as I haven't found a picture of the P317 variant without the cooler.

EDIT:

Since it's the same PCB as the 6800GS, here we can clearly see it's the memory IC nearest to the PCIe slot:
https://theretroweb.com/expansioncards/s/msi-nx6800gs

Cheers.
Yes, it's identical to that.
I'll try a reflow when I get round to it, but I guess there's no way to tell that it is definitely the ram and not gpu. Both look good from the side but that means nothing.

Tempted to recap it.
Looks like Panasonic bought the original brand, but being 20yrs, the exact models don't exist.
Stick with Panasonic? Or do you have another recommendation?

The 533 on the caps in the picture is different to mine. Date code? 2005 w33?

Anyone happen to have ancient knowledge of the nvidia bios on these?

I suspect that it is possible to disable memory banks, but the straps are... Not the way to go.

I think if it exists it's in the memory reset script.

I've looked at the Linux nouveau driver, and envytools, and so far nothing stood out.