From what you say, I would first consider the soldering.

It might be an option to "bake" the card, if you know what I mean.

Baking is a horrible thing, NEVER EVER do that please, it pretty much ruins capacitors, connectors, excessive heat load on the PCB...

I have a hot-air soldering station but messing with BGA chips is out of my league, I don't know for how long and how much to heat the chip. I once tried to "reflow" a GeForce3 and it ended up with the solder balls melting completely, had to have it reballed.

Jofo wrote on 2024-05-17, 12:39:

Baking is a horrible thing, NEVER EVER do that please, it pretty much ruins capacitors, connectors, excessive heat load on the PCB...

I have a hot-air soldering station but messing with BGA chips is out of my league, I don't know for how long and how much to heat the chip. I once tried to "reflow" a GeForce3 and it ended up with the solder balls melting completely, had to have it reballed.

It worked better for me than your method worked for you. 😉

ux-3 wrote on 2024-05-17, 13:46:

Jofo wrote on 2024-05-17, 12:39:

Baking is a horrible thing, NEVER EVER do that please, it pretty much ruins capacitors, connectors, excessive heat load on the PCB...

I have a hot-air soldering station but messing with BGA chips is out of my league, I don't know for how long and how much to heat the chip. I once tried to "reflow" a GeForce3 and it ended up with the solder balls melting completely, had to have it reballed.

It worked better for me than your method worked for you. 😉

Think for a minute about what you are saying, and the act of “baking” the card. If you are heating the card up enough to melt solder then any SMD components on the underside are at risk of detaching. If you aren’t heating the card up enough to melt solder, then it achieves nothing permanent.

Baking electronics is a fool’s errand.

I "Baked" a card I didn't care about, and it did make it work again, but only for a relatively short amount of time; it cured the symptom rather than the underlying problem. And that seems (from what I've read) to be the best case scenario outcome, a temporary fix at best.

But if it works again, you know it is a solder issue.
The initial plan may call for the destruction of two cards without knowing the cause. Not sure if that is better.

But do as you please, it is your stuff.

It requires good equipment and practice for bga reball . Also preheating station is a must .

Back to the topic guys - the question is, would a chip from a Quadro FX1300 work in 5950 Ultra? Or is it somehow hard-wired as Quadro?

Description of the issue sounds much more like memory issue and not chip issue.

Transistors inside die almost never break and its either:
1. connection between chip package and cards BCB
2. connections inside chip package and die itself
In both cases card would probably work and just have random crashes (especially due to vibrations) or not work at all. Eventually crash after some time as chip heats up.
Garbled screen... almost definitely memory chip is faulty. These won't be good from FX1300 - you need to find different donor.

Replacing chip from a donor only makes sense when you exclude memory and first case - by doing normal reballing. Then if it breaks soon then its probably not connection to PCB - if repair job was done correctly and it was proper reballing and not guy with heat gun pretending they are BGA repair specialists...

Jofo wrote on 2024-07-04, 17:41:

Back to the topic guys - the question is, would a chip from a Quadro FX1300 work in 5950 Ultra? Or is it somehow hard-wired as Quadro?

No, you can use it. The GPUs on both cards are identical, only the heatspreader printing is different. Same with FX1000 and 5800 Ultra for example.
The difference between a Quadro and a GeForce happens 1) in the BIOS and 2) in the straps (which can be both hard- and softmodded). But if you just plop the Quadro GPU on another PCB, it'll think it has always been a GeForce. There's a slight chance the binning is different and the Quadro GPU may not clock as high, but the chances for that are pretty low imo. If you try to OC the Quadro to check, remember the 5950 cards have higher voltages (1,3 to 1,5V I think it was).

I have just reballed two NV38 and one NV35 GPU and sadly, none of those were successful. Likely more PCB damage or another culprit. Try to make sure that this is actually the problem because it's a bit of a last resort. Even if you're really careful, 20 year old GPUs don't like to be soldered. It can work but I'd say it's a 50:50 at best. Also I'd advise to delid the new GPU before soldering, so you need less heat. I just soldered a 4600 GPU with 180°C bottom heat and 230°C top heat for example which is comparably very low.

Were the artifacts by any chance looking like this? https://www.youtube.com/watch?v=HrtDtSHQ97Q&ab_channel=NDG93
This is from my Leadtek 5950 Ultra before i took the GPU off.

Jofo wrote on 2024-05-17, 12:29:

The card normally boots into Windows, no artifacts, but when running 3DMark - all hell breaks loose on the screen.

The thermal interface between the GPU die and the heat spreader could be damaged. I had that happen on my GeForce FX5900. I fixed it by delidding it with a razorblade and removing the stock thermal interface material. Now i am running it with a aftermarket cooler directly mounted on the die.

But before you start delidding yours, check the GPU temperature first. If I recall correctly you can find it in the Nvidia control panel. Before delidding mine idled at 80-90C and crashed in 3D at 115C. Now, with the new cooler, it doesn't even reach 60C.

@Masterchief79 - thanks, this is very valuable info. I currently don't own a BGA rework station but probably not a fancy one since I am just a hobby user, so would train a lot on other chips before attempting this one. Good to know that it's doable, I would first try to run that Quadro on Ultra clocks (for core) if it's stable.

However, a rather strange thing happened. I wanted to shoot a video of the artifacts since you mentioned it, so I've plugged the card into the PC, same config as before. Just noting that the card was in service where I was told that the chip is most probably damaged and I have not touched the card since then. I turned on the PC , run 3DMark03 and - it went just fine. There was only like a half of second artifact in the beginning of the Ragtroll scene (not even on the run from the video below). Then I run 3DMark01 and also - only on the spinning horse scene had something off for like 0.5s but otherwise fine. I run Aquamark3 completely without problems. However, 3DMark05 was a mess, shorty after start and then the FPS was really low - all hell broke loose on the screen, 1FPS broken-image-spree. But even when I finished the benchmarks and fired up Max Payne, the game run without trouble. I cannot explain this... Before that it was 3DMark01 that caused massive artifacting, now it ran just fine.

https://www.youtube.com/watch?v=6vZHIpnqIVE

paradigital wrote on 2024-05-17, 15:59:

ux-3 wrote on 2024-05-17, 13:46:

Jofo wrote on 2024-05-17, 12:39:

Baking is a horrible thing, NEVER EVER do that please, it pretty much ruins capacitors, connectors, excessive heat load on the PCB...

I have a hot-air soldering station but messing with BGA chips is out of my league, I don't know for how long and how much to heat the chip. I once tried to "reflow" a GeForce3 and it ended up with the solder balls melting completely, had to have it reballed.

It worked better for me than your method worked for you. 😉

Think for a minute about what you are saying, and the act of “baking” the card. If you are heating the card up enough to melt solder then any SMD components on the underside are at risk of detaching. If you aren’t heating the card up enough to melt solder, then it achieves nothing permanent.

Baking electronics is a fool’s errand.

Agreed, if you took the time to remove heat sensitive parts, and set it in a bed of sand or something to support the under components, maybe.

But chances are that it isn’t going to go well.

A bga job the likes of a 5950 series is not a task for an amateur. You need some serious skill and equipment to do that in any meaningful way that might work. A spot heating of the gpu, will probably also fail, but stand a better chance, especially if you have access to an x-ray to verify your work.

@Jofo
If artifacts don't happen consistently, they might be temperature-dependent. For example, if you have a hairline crack in a GPU or memory solder ball, it can temporarily make contact again when the card is hot. It can also be orientation-dependent (if you have the card in a case or sitting upright on a benchtable), because of the mechanical stress put on the PCB by gravity and the heavy cooler, combined with heat.
Unfortunately this still means your card has some kind of defect and to get rid of it, you'll most definitely have to at least reflow something.

If you can run Nvidia Mods/Mats, you can pinpoint the problematic memory chip. I would suggest that next. Version 173-48 works with 5900 Ultra: https://drive.google.com/file/d/1WV5Z5lmM_A6g … ?usp=drive_link
Alternatively Version 70/75: https://drive.google.com/drive/folders/1pFuLL … kJDv4TiJKE_GWNw

1. Make USB Stick with Rufus and Select FreeDos
2. unzip files and copy folder 173-48 to USB Stick
3. Boot from it
4. type "cd 173-48"
5. type "mats -e 10"
6. post results

should look something like this:

Masterchief79 wrote on 2024-07-10, 19:03:

Version 173-48 works with 5900 Ultra: https://drive.google.com/file/d/1WV5Z5lmM_A6g … ?usp=drive_link

Unfortunately, to download from this link, you need to register. You could not open free access to this file or upload it again to any other open file sharing service that does not require registration, such as YandexDisk, MediaFire.com or Upload.ee (Or any other one that doesn't require registration for downloading). Not everyone has a Google account. Thank you.

DoZator wrote on 2024-07-11, 10:06:

Not everyone has a Google account. Thank you.

I didn't think that was possible. Here you go: https://www.mediafire.com/file/7uj9ogssvz8bva … 173-48.zip/file

paradigital wrote on 2024-05-17, 15:59:

Think for a minute about what you are saying, and the act of “baking” the card. If you are heating the card up enough to melt solder then any SMD components on the underside are at risk of detaching. If you aren’t heating the card up enough to melt solder, then it achieves nothing permanent.

Baking electronics is a fool’s errand.

No, none of the components on the back (bottom) will fall off, even if heated to proper solder melting temperatures. The reason is that solder has somewhat similar liquid properties to water in that it will have a holding force. For all of the small SMD components, rest assured that none of them will fall off. And even for heavier stuff like RAM chips, they will still keep in place.
The hardest part about doing rework on board is how the board is propped up / supported on the bottom. Not enough supports can result in the PCB to warp (this is especially the case with large PCBs - i.e. motherboards.) Also, no part of what's propping the board should exert pressure on any component, so that the risk of it getting moved from that tension is minimized.

However, I do agree that baking is indeed a fool's errand... well, to a point. Low-temperature baking (under 110C) is actually useful for drying / removing the moisture from old PCBs before sending them in for re-work. If the PCB has ever lived in a more humid environment, then baking is almost crucial prior to rework. Otherwise, the risk of "popcorning" the PCB increases a lot.
But apart from that, using baking as a means to "reflow" a board is indeed a fool's errand. You're better off with a heat gun and a temperature probe in hand (and I can speak from experience here, as I still have working Radeon 9700 video cards back from when I was trying to learn rework, but had nothing more than a heatgun and a type-K thermocouple + meter.)

That said, you really don't need that much "special" equipment to do rework. You just need a lot of experience and also a good understanding of thermal flow. Of course, both of these will come easier (and quicker) if you do have the proper equipment to begin with. But even if you don't, it's still possible to get there with a lot of practice.

In my case, I don't have a BGA machine or any sort of proper hot air rework station. But I did learn on one many years back in a console repair shop (~2011 - i.e. peak Xbox 360 and PS3 failure era.) And once I got that knowledge, I quickly figured that just about any gas or electric stove top (induction type excluded, obviously) coupled with a cheap hot air station and my type-K meter will work just as well as a full rework station for at least 75% of the jobs. The previous place where I lived, we had natural gas and a gas stove top. That was, by far, *the* best rework equipment I've used. You might think that an open flame will damage stuff on a PCB, but it's actually a very safe way to heat up PCBs. Only key to that is to have the PCB placed at the right height above the flame (i.e. not touching it.) A flame emits both IR heat and creates hot air. The former heats the board through conduction, and the latter through convection. So the heating is actually very even on all of the components above the flame, which is good. At the same time, the flame is a fairly spot-concentrated heat source, so it's great for heating only the part of the PCB you want and won't overheat components on the rest of the PCB (e.g. plastic connectors and etc.) With a hot air source from the top, I can remove quite a lot of BGA stuff. A few months back, I removed a shorted socket A Duron CPU from a K7SEM board. No damage at all to the board and CPU came off nicely with all of its pins out.

Anyways, the point is, you can do quite a lot of stuff even with more limited equipment. Just make sure to practice on something that isn't important or better yet completely scrap parts. Once you get good at it, then move onto the more serious stuff.

I have some new insight on this topic. Allow me to tell the whole story:
I bought this gorgeous Gigabyte FX5950 Ultra recently as defective. I repaired it and it worked for a while. After a few days of gaming Max Payne 2 (awesome game btw), it greeted me with picture artifacts again. A Mats/Mods test revealed that the exact same memory channel and bit were faulty as before, although I reballed the memory chip.

This means the broken solder joint was underneath the GPU all along (top right corner, B1 ramchip, bit B11) and had only temporarily been "repaired" by my BGA work on the memory chip.

I reballed the GPU and unfortunately killed it in the process. With these old GPUs, there is like a 20-40% chance that you do everything correctly and they still roll over dead.

Then I tried two more donor-GPUs that I pulled from two more (already very defective) 5950 Ultras by Leadtek and Asus respectively. After reballing and soldering, the Leadtek GPU was only putting out absolute pixel salad on my previously working card. With the second GPU, my Gigabyte card was technically working again and the memory errors were gone! Sadly, the picture was completely blue, and the red and green channels were not working. I traced this to be a GPU-internal defect - yet again.

I had one more thing to try: Steal the NV38 GPU of my (working) Quadro FX1300. This is the chip transplant this thread was originally asking about. You can pick up those cards for like 20€ if you're lucky, which is the absolute cheapest way to get your hands on a NV38.

This is the quadro board I stole it off of:

GPU reballed and soldered onto my Gigabyte FX5950U:

This resistor is placed differently on Quadro GPUs and most likely decides whether the GPU is recognized as a NV38 (5950U) or NV38GL (FX1300). You have to delid the GPU to get to it, but I wouldn't recommend soldering GPUs with IHS anyway, so might as well do it in the process.

In terms of clockspeeds: You can OC the 350MHz Quadro card before transplanting and test what clocks it will do. You're not guaranteed to have a GPU that can run the 5950U-spec. But keep in mind the Quadros only run the GPU at 1.35V. My Quadro did 425MHz which is quite good (two other cards crashed at 400MHz). A 5900U also runs at 425MHz but uses 1,45V. On a 5950U, that's 475MHz@1.55V. Most NV38 chips should be able to do that.
If your FX1300 can run around 400MHz without mods, you should be good.

Now, my card works again. I hope it stays that way, it was a busy few days. ^^

Masterchief79 wrote on 2024-08-19, 09:21:

I have some new insight on this topic. Allow me to tell the whole story: I bought this gorgeous Gigabyte FX5950 Ultra recently a […]
Show full quote

I have some new insight on this topic. Allow me to tell the whole story:
I bought this gorgeous Gigabyte FX5950 Ultra recently as defective. I repaired it and it worked for a while. After a few days of gaming Max Payne 2 (awesome game btw), it greeted me with picture artifacts again. A Mats/Mods test revealed that the exact same memory channel and bit were faulty as before, although I reballed the memory chip.

The attachment AGC_20240717_112817285.jpg is no longer available

This means the broken solder joint was underneath the GPU all along (top right corner, B1 ramchip, bit B11) and had only temporarily been "repaired" by my BGA work on the memory chip.

The attachment 20240408_124757.jpg is no longer available

I reballed the GPU and unfortunately killed it in the process. With these old GPUs, there is like a 20-40% chance that you do everything correctly and they still roll over dead.

Then I tried two more donor-GPUs that I pulled from two more (already very defective) 5950 Ultras by Leadtek and Asus respectively. After reballing and soldering, the Leadtek GPU was only putting out absolute pixel salad on my previously working card. With the second GPU, my Gigabyte card was technically working again and the memory errors were gone! Sadly, the picture was completely blue, and the red and green channels were not working. I traced this to be a GPU-internal defect - yet again.

I had one more thing to try: Steal the NV38 GPU of my (working) Quadro FX1300. This is the chip transplant this thread was originally asking about. You can pick up those cards for like 20€ if you're lucky, which is the absolute cheapest way to get your hands on a NV38.

This is the quadro board I stole it off of:

The attachment AGC_20240819_110805509_2.jpg is no longer available

GPU reballed and soldered onto my Gigabyte FX5950U:

The attachment AGC_20240819_102836347.jpg is no longer available

This resistor is placed differently on Quadro GPUs and most likely decides whether the GPU is recognized as a NV38 (5950U) or NV38GL (FX1300). You have to delid the GPU to get to it, but I wouldn't recommend soldering GPUs with IHS anyway, so might as well do it in the process.

The attachment IMG_20240819_103216101_edited.jpg is no longer available

In terms of clockspeeds: You can OC the 350MHz Quadro card before transplanting and test what clocks it will do. You're not guaranteed to have a GPU that can run the 5950U-spec. But keep in mind the Quadros only run the GPU at 1.35V. My Quadro did 425MHz which is quite good (two other cards crashed at 400MHz). A 5900U also runs at 425MHz but uses 1,45V. On a 5950U, that's 475MHz@1.55V. Most NV38 chips should be able to do that.
If your FX1300 can run around 400MHz without mods, you should be good.

Now, my card works again. I hope it stays that way, it was a busy few days. ^^

Wow, congratulations!
Thanks for detailing the process and confirming that the FX1300 is a good donor for the FX5950. Great use for the card 😀 It's a shame they used those bridge chips just to get PCIe cards out the door back then and that greatly limits their utility today.
BTW I was going to ask what equipment you use but can see from one of your youtube videos that you've got a hotplate and a decent hot air station, it must take some experience to be able to swap over chips like this, I wish I was better with BGA.
For temperatures on the FX series, do you use temp ranges for leaded solder or were they using lead-free by this point?