Here is the mats output

That should be a memory pin on the edge of the GPU. I just checked the schematic and FB-A Data pin 35 is at AW38 which I think is a corner of the chip since there are no column letters used past AW and 38 is one off the lowest row I can see which is 39. So your failed memory pin should be one corner of the chip which might happen either from where a heatsink was overtightened (which I've done before) or maybe just failure from age/mechanical stress.
Or maybe it's physical damage to a trace - the AK38 pin should be in the bottom right of the chip by the end of the AGP slot. There's a mark on one of the traces down there by the weird little corner drill hole. Is that dust or a damaged trace?

Could check for a bad RAM chip, having a schematic is again so nice and the FX5900 cards are mostly reference designs. The one with FBA data 35 is U11. But in my experience individual dead pins aren't often caused by bad RAM.
Wanna see the schematic for yourself? I got it here and it seems complete: https://elektrotanya.com/nvidia_35p177_nv35_b … f/download.html

I was doing something similar trying to work out what to do with my FX5600 cards with bad RAM pins: Re: What retro activity did you get up to today?

Don't reflow just yet but if it comes to that then hopefully someone skilled at BGA soldering could give some suggestions too, it seems these 256-bit NV GPUs are hard to reflow / reball from what masterchief79's experience shows: Re: Broken GeForce FX5950 Ultra - chip transplant

Thermalwrong wrote on 2024-09-16, 01:38:

That should be a memory pin on the edge of the GPU. I just checked the schematic and FB-A Data pin 35 is at AW38 which I think i […]
Show full quote

That should be a memory pin on the edge of the GPU. I just checked the schematic and FB-A Data pin 35 is at AW38 which I think is a corner of the chip since there are no column letters used past AW and 38 is one off the lowest row I can see which is 39. So your failed memory pin should be one corner of the chip which might happen either from where a heatsink was overtightened (which I've done before) or maybe just failure from age/mechanical stress.
Or maybe it's physical damage to a trace - the AK38 pin should be in the bottom right of the chip by the end of the AGP slot. There's a mark on one of the traces down there by the weird little corner drill hole. Is that dust or a damaged trace?

Could check for a bad RAM chip, having a schematic is again so nice and the FX5900 cards are mostly reference designs. The one with FBA data 35 is U11. But in my experience individual dead pins aren't often caused by bad RAM.
Wanna see the schematic for yourself? I got it here and it seems complete: https://elektrotanya.com/nvidia_35p177_nv35_b … f/download.html

I was doing something similar trying to work out what to do with my FX5600 cards with bad RAM pins: Re: What retro activity did you get up to today?

Don't reflow just yet but if it comes to that then hopefully someone skilled at BGA soldering could give some suggestions too, it seems these 256-bit NV GPUs are hard to reflow / reball from what masterchief79's experience shows: Re: Broken GeForce FX5950 Ultra - chip transplant

Thank you for the detailed reply.

I think that is just dust or a surface mark but I'll have a look under the microscope to be sure.

I have a cheap hot air station so I can try a reflow on the core, although my experience in doing reflows is limited. I've also never successfully reballed.

Hopefully it's just a bad solder joint rather than a broken pad under the core or memory because I don't have the skill to repair that.

byte_76 wrote on 2024-09-17, 12:42:

Thank you for the detailed reply. […]
Show full quote

Thermalwrong wrote on 2024-09-16, 01:38:

That should be a memory pin on the edge of the GPU. I just checked the schematic and FB-A Data pin 35 is at AW38 which I think i […]
Show full quote

That should be a memory pin on the edge of the GPU. I just checked the schematic and FB-A Data pin 35 is at AW38 which I think is a corner of the chip since there are no column letters used past AW and 38 is one off the lowest row I can see which is 39. So your failed memory pin should be one corner of the chip which might happen either from where a heatsink was overtightened (which I've done before) or maybe just failure from age/mechanical stress.
Or maybe it's physical damage to a trace - the AK38 pin should be in the bottom right of the chip by the end of the AGP slot. There's a mark on one of the traces down there by the weird little corner drill hole. Is that dust or a damaged trace?

Could check for a bad RAM chip, having a schematic is again so nice and the FX5900 cards are mostly reference designs. The one with FBA data 35 is U11. But in my experience individual dead pins aren't often caused by bad RAM.
Wanna see the schematic for yourself? I got it here and it seems complete: https://elektrotanya.com/nvidia_35p177_nv35_b … f/download.html

I was doing something similar trying to work out what to do with my FX5600 cards with bad RAM pins: Re: What retro activity did you get up to today?

Don't reflow just yet but if it comes to that then hopefully someone skilled at BGA soldering could give some suggestions too, it seems these 256-bit NV GPUs are hard to reflow / reball from what masterchief79's experience shows: Re: Broken GeForce FX5950 Ultra - chip transplant

Thank you for the detailed reply.

I think that is just dust or a surface mark but I'll have a look under the microscope to be sure.

I have a cheap hot air station so I can try a reflow on the core, although my experience in doing reflows is limited. I've also never successfully reballed.

Hopefully it's just a bad solder joint rather than a broken pad under the core or memory because I don't have the skill to repair that.

I was thinking that perhaps you could try removing U11 first to see if the pad is good.
I'm not any good at BGA reballing yet either and BGA reflow is still not something I like doing, keep picking up too many side projects to work on the BGA stuff 😀

I've checked and confirmed that it was just some lint on the board. All tracks from the memory to the core look good, I cannot see any damage.

Looking at it from an angle where I can see the solder balls under the core, it looks like the core is slightly lower on the bottom right corner. As if the balls are slightly more flat.

I'm not comfortable with removing module U11 because I don't have the skill or the stencil to reball the module.

I can try a reflow of the core.

I've tried a reflow on the GPU core but it made no difference at all. Still have an error with A35.

I removed the memory module U11 some time ago and did not find any issues so I reballed and resoldered the module.

Yesterday I removed the core and this time I found the cause of the issues.
On the board there are multiple pads under the core in the bottom left and right corners which have been ripped, with at least 3 of those being connected to traces.

Not sure if I will be able to fix this as I've never done repairs of pads before and they are so small.

So this is my messy amateur attempt at pad repair.

I scratched off a small section of the trace leading up to the pads and then used a tiny amount of super glue on the end of a toothpick dabbed onto the pad holes on the board in an attempt to hold the pads in place. I then soldered the tails to the traces.
This process was very challenging and I ended up using many pads to repeat the process over and over again because the pads kept moving or springing away or sticking to the soldering iron while trying to hold them down and solder the tails.

Anyway, I managed to get it to the point where it might work and all of the solder joints have been covered with solder mask

I used my multimeter to measure from the pads to ground and each pad gives me a resistance reading and none of the pads are shorted to each other.

Next step will be to solder back the core but I'm waiting for a reballing stencil to arrive and then I'l prepare the core and solder it back.

Good job, but I have a remark. So called "super glue" (cyanoacrylate) does not sustain high temperatures. Better off use an UV-curing solder mask on the new pad's "tails" to fix them. And hope that the "super glue" won't cause any problems when soldering the BGA chip, 'cause it will turn into a burnt black mass.

The super glue was only used under the pads to hold them in place so that I could solder the tails because the pads kept moving and I was struggling to hold them steady.

I did use green UV solder mask on the tails after they were soldered. You can see the mask in the pics.

If it doesn’t work, I guess it’s too bad. I really struggled with this for many hours but at least I tried.

Hello byte_76,

I have a similar problem with my Geforce 5900 Ultra 256MB.
The Computer is starting normally, but when I load the Graphics-Driver in Windows XP I get a black screen. If I bend the running card a little, I get yellow stripes on the screen.
I can't find a working version of MATS. Do you have a link to your used version? Maybe by PM? Thank you.

amai wrote on 2025-01-28, 14:11:

Hello byte_76, […]
Show full quote

Hello byte_76,

I have a similar problem with my Geforce 5900 Ultra 256MB.
The Computer is starting normally, but when I load the Graphics-Driver in Windows XP I get a black screen. If I bend the running card a little, I get yellow stripes on the screen.
I can't find a working version of MATS. Do you have a link to your used version? Maybe by PM? Thank you.

It’s so hard to find a working FX5900 these days.

Here we discussed mods near the bottom of page 5:

Any late era Win9x AGP GPUs with good backward compatibility?

My second FX5900 had a dead core and since I only recently acquired this card, I don’t know what caused the failure.

I managed to get a replacement core and soldered it to the board.

After soldering, I used my AGP tester to check connections and found that B60 (data line AD7) indicates a short.

Is it most likely that I got bridged balls under the core, or could there be a bad component on the board that is shorting the data line to ground? (Maybe also the cause of the original dead core)
Maybe one of the caps on the board behind the core?

I’ve tried testing the caps on the board but they all indicate short at the moment, even with an LCR meter.

I’m not sure how to read the schematic linked earlier in the thread. Any ideas which cap might be linked to B60, if any?