It looks like C and D chips have issues. Next step is to find where they are located on the board and if replacements are available.

It appears the memory chips are "HYB18H512321AF-14". Unfortunately, no one sells this exact model anymore.

Options are ;
- Try the identical model but slower speed "HYB18H512321AF-13". https://www.aliexpress.com/item/1005008640348104.html (this would involve dropping the memory clock to 675 MHz).
- Try a slightly different model with identical specs (I think) https://www.aliexpress.com/item/1005007174051898.html
- Find other cards that used "HYB18H512321AF-14" and harvest from that. Maybe a 7800GT PCIe ?

I think the memory order goes like this for older cards. I've highlighted the bad chips. The MATS log says bank 0 (EB0) and bank 1 (EB1) of D has errors as well as C0. In this situation, that is referring to a single chip. On something like a 7800GTX bank 0 is on the front of the card and bank 1 is on the back of the card, two different chips.

Since D1 has millions of errors, a reflow of the chip is probably not going to work. A reflow of C0 might work since there aren't too many errors. In any case, I'm going to replace both chips since it's such a rare card.

I replaced D1 and reflowed C0 but now there are more memory errors. I think the next step is going to be reflowing or reballing the core.

1  LANE    EB0 READ ERR EB0 WRT ERR  EB1 READ ERR EB1 WRT ERR  UNKNOWN ERR 
2---------- ------------ ------------ ------------ ------------ ------------
3FBA[31: 0]           0            0            0            0            0
4FBA[63:32]           0            0            0            0            0
5FBB[31: 0]           0            0            0            0            0
6FBB[63:32]           0            0            0            0            0
7FBC[31: 0]      753975     24244283       945925     24040487            0
8FBC[63:32]           0            0            0            0            0
9FBD[31: 0]     7093405     18112739      6987330     18218174            0
10FBD[63:32]          98     16817461          194     16816754            0

Voltages?
Tried underclocking?
Checked smd caps nearby for shorts?

Mats is brilliant, but it can only point in the general direction of the fault.
It doesn't mean necessarily that the first assumption is correct

myne wrote on 2026-01-03, 05:33:

Voltages? Tried underclocking? Checked smd caps nearby for shorts? […]
Show full quote

Voltages?
Tried underclocking?
Checked smd caps nearby for shorts?

Mats is brilliant, but it can only point in the general direction of the fault.
It doesn't mean necessarily that the first assumption is correct

Apologies for the late reply. Had to jump to some other retro duties. I've had some time to come back to this . Can't find any shorts. Voltages appear to be OK although I don't have the schematics for this card to double check.

I've installed a new core...

and replaced the memory using memory from a harvested card (https://theretroweb.com/expansioncards/s/ati- … -xt-pci-e-256mb) plus reballed as these chips can't be purchased anywhere. I confirmed the memory on the card was working perfectly before harvesting.

Unfortunately, exact same issue exists. So there must be a component on the board that's dead that I can't identify. I've been told by someone who repairs a lot of AGP cards that these 7800GS's are almost impossible to repair. A bit disappointing considering the amount of time I've put into this card but as they saying goes, you win some and you lose some...

Here are some resistance measurements in case anyone needs them for their card in the future.

I'd be suspecting the bridge chip if everything else checks out

It doesn't look like you protected it from the heat either

Scratch that. I just read through the thread again

OK. So it works, but one memory channel is bad. Yes?
If you have an appropriate donor, swap all the smd components around that chip.

These traces may be damaged by screws, handling, and time

The red looks sus too

myne wrote on 2026-03-08, 03:17:

I'd be suspecting the bridge chip if everything else checks out […]
Show full quote

I'd be suspecting the bridge chip if everything else checks out

It doesn't look like you protected it from the heat either

Scratch that. I just read through the thread again

OK. So it works, but one memory channel is bad. Yes?
If you have an appropriate donor, swap all the smd components around that chip.

2 mem chips are dead. D1 and C0.

Unfortunately I don’t have a donor board, but yes swapping all the components would be a good idea.

I’ll have a closer look at that screw hole. Thanks for the heads up.

Look for 8 and 9x00s with the same memory size and bus width.
They're dirt cheap and given the way tech works, in all probability they will have the same bga footprint and slightly better specs.
I doubt there will be any compatibility issues running faster chips with slower timings.

Eg https://www.techpowerup.com/gpu-specs/geforce-9600-gt.c206

They're cheap. Even sticking with gainward for the probability that they use the same bga footprint

https://www.ebay.com/itm/397462059319?_skw=Ga … atedSearch=true