Try the card in a Windows 2k or XP system (I'd lean towards XP, since it will likely already have the drivers for it from the start) and see where that takes you.
I suspect the card has a BGA issue or a chip-level issue.

This card came from your recycler buddy, am I right? If so, who know what could have happened to it prior to you getting it. Maybe it sat at the bottom of a scrap bin and got hit with who knows what. Or maybe it was just "ripped out" from the slot, possibly bending the PCB in damaging ways. Who knows. Recyclers tend to have satanic hands and methods, IMO. I almost want to say they get satisfaction from torturing old hardware like this (well I said it now, didn't I 😁 ).

Most likely a dead card if drivers installation leads to severe problems. I also have a geforce 256 and it is working fine however it was the 3 attempt to get one in working condition!

momaka wrote on 2025-07-18, 07:15:

This card came from your recycler buddy, am I right? If so, who know what could have happened to it prior to you getting it. Maybe it sat at the bottom of a scrap bin and got hit with who knows what. Or maybe it was just "ripped out" from the slot, possibly bending the PCB in damaging ways. Who knows. Recyclers tend to have satanic hands and methods, IMO. I almost want to say they get satisfaction from torturing old hardware like this (well I said it now, didn't I 😁 ).

Not this one. It had been taken proper care of. As for BGA - I'vre tsken a hot air station (Yihua 862BD+) to the main GPU - 365*C, 80% fan speed, small-ish nozzle, for no longer than 2 minutes tops. No change in the artefacts. VMTCE recorded 100.000 errors in the last run I've did yesterday (2 or so hours after the photo I took in the "What retro activity did you get up to" thread).

As for 2k/XP... see below.

(and funny enough - a 5900XT from the recycler still works despite having been through conditions that were... unorthodox, to say the least, but this one, that had been obviously taken care of, exhibits these issues.)

Looks like memory to me, perhaps dry joints on one or two of the memory ICs, or dead ones.

Trashbytes wrote on 2025-07-18, 08:13:

Looks like memory to me, perhaps dry joints on one or two of the memory ICs, or dead ones.

All memory joints were redone 4 times in a row with no change.

Even modern gpus behave the same.
Code 43 fine in 2d.
And yes, it's mostly memory.

If you can find a truly ancient version of mats mods that will help.

Can't hurt to check voltages though. Maybe the memory power is dead. Also look for dead shorts on caps, and opens on resistors.

myne wrote on 2025-07-18, 10:15:

Even modern gpus behave the same. Code 43 fine in 2d. And yes, it's mostly memory. […]
Show full quote

Even modern gpus behave the same.
Code 43 fine in 2d.
And yes, it's mostly memory.

If you can find a truly ancient version of mats mods that will help.

Can't hurt to check voltages though. Maybe the memory power is dead. Also look for dead shorts on caps, and opens on resistors.

So far I've replaced the caps (tantalum, 2x100uF and a 150uF) and no change. The only code I got from it under 9x is 24, and that is in 2D.

Ok, but there are many more than just the big caps.

Smd components go bad eventually too - which can include the ram itself.

High res pics of the board might help

Pics

R111 looks damaged
C133 too

C17, c24, c26, c72, 73

Hard to tell from a pic, but gives you somewhere to start looking

myne wrote on 2025-07-18, 11:17:

R111 looks damaged C133 too […]
Show full quote

R111 looks damaged
C133 too

C17, c24, c26, c72, 73

Hard to tell from a pic, but gives you somewhere to start looking

C133 is intact so far. I replaced R111 to be sure, and none of the caps listed below are shorted or damaged.

There's not many thing that could fail in significant manner, so my guess would be an internal damage to one of the memory chips. Perhaps a genuine BGA damage due to shipping or something? But I find it more hard to believe.

@TS: Did you try different PSUs? Maybe it got a weak or broken 3,3V line.

I encountered something similar 25 years ago using a GeForce DDR on an Asus P2L97 motherboard, black screen with driver installed. The motherboard used an onboard regulator for generating 3,3V from 5V that wasn't powerful enough.

Different PSUs were used:

- bequiet 300W (FSP)
- Altis 450W (HEC/Compucase)
- Allied AL-A350ATX (Solytech, recapped)
- Spire SP-600W-ATX (Topower, recapped)
- ISO-500PP (CWT, recapped)

No difference with any of those. Same behaviour also happens across several mainboards, from as old as Gigabyte BX2000 and P3B-F to as new as Gigabyte 8IPE1000 (i865) and AOpen i855GMEm-LFS.

Right, but did you check the voltages on the card?

And did you look for mats/mods?

This works on gf3. Fair odds it will go back to gf256.

GeForce3 - mixing memory modules

Hmm...
Can you get to the card's BIOS and edit or re-flash it?
I wonder if using lower memory and/or core clocks will change anything.
Normally SDRAM is relatively tough and rarely fails. But I can see your card has a rather cheap-looking RAM brand (who the heck is "ESMT"?) Perhaps a RAM IC is indeed failing somewhere. But before you tear those out, I think it might also be worth checking the clock crystals on the card. It's very unlikely they are failing... but if one is, perhaps that's messing up the reference clock of the GPU and/or RAM? If you have an o-scope or frequency counter capable of measuring that high, check these just in case.

myne wrote on 2025-07-18, 10:15:

Can't hurt to check voltages though. Maybe the memory power is dead.

Agreed.
While it's very unlikely for some voltage reg to have a drifting low/high voltage, it IS worth to check indeed. It's rare, but voltage regulators can "semi-fail" - still regulate somewhat, but not to the exact voltage they are supposed to. Had that happen with a 7805 reg in an audio amp. The 7805 was getting a steady 9V on the input, but on the output, it could barely manage 4.65... and it wasn't even close to over-loaded.

myne wrote on 2025-07-18, 10:15:

Also look for dead shorts on caps, and opens on resistors.

+1
Saved a Radeon HD6850 this way some years ago. Had a shorted ceramic cap close to a memory chip. The shorted cap was not for V_ddq or V_tt filtering, but something else (in series with a resistor and going to one of the memory chips.) Replaced and got the card back up and running (no more patterned artifacts.)

Kamerat wrote on 2025-07-21, 11:40:

@TS: Did you try different PSUs? Maybe it got a weak or broken 3,3V line.

A "broken" 3.3V rail would halt the whole system before POST even occurs.
A weak (sagging) 3.3V rail can cause these issues... but I figured PcBytes has already tried different PSUs since he has quite a few. And in general, these old video cards can't really generate enough of a load on the 3.3V rail to make a difference to the PSU that much. If/when they do, usually the whole PC crashes. Weak/insufficient power delivery from the PSU almost never causes graphics artifacts, IME.

MATS result.

I'm going out on a limb and guessing Noname sdram is genuinely faulty.
I got a bunch of normal sd sticks a couple years ago and about half of them had something wrong - from dead dead to bad bits to wouldn't run at rated speed.

One of them is good. If you confirm the voltages are right, you could try swapping them around and see if the errors/good move too.

Swap the first 2 first. If 1 comes good and 2 goes bad, the core is likely good.

Iirc the memory goes anticlockwise from the bottom. But you'll have to check.

Actually, I somehow didn't notice there are "2 banks"
One "bank", probably on the front are all fine.

I'd be curious to see what happens when you remove one at first (then run mats), and then all of the rear mem modules (then run mats).
I would guess that it would become a perfectly functional 16mb card.

It's possible there's a problem with the bank select signal which branches off from the one working chip. It may only be one line.

But this is pure educated guessing.

Couldn't hurt to check the ram datasheet and trace the bank select pin between each chip.
It could just be a dud resistor.