TrashPanda wrote on 2023-03-15, 10:45:

That type of fix really only buys you a bit of extra time, eventually the solder will crack again. Its pretty much the same principal as people baking their 8800GT cards ..it buys you enough time to replace the card but it will eventually die again and normally far quicker than you expect.

I agree. I've had very good luck with "proper" station work though. PC which I use for old games has NV 4400ti which was repaired and has been 1000+ hours on in 2 years time.

Baking method heats whole card which isn't something I wouldn't say worth it. ie. G80 die is the problem and it's going to die 100% again.

timsdf wrote on 2023-03-15, 10:56:

TrashPanda wrote on 2023-03-15, 10:45:

That type of fix really only buys you a bit of extra time, eventually the solder will crack again. Its pretty much the same principal as people baking their 8800GT cards ..it buys you enough time to replace the card but it will eventually die again and normally far quicker than you expect.

I agree. I've had very good luck with "proper" station work though. PC which I use for old games has NV 4400ti which was repaired and has been 1000+ hours on in 2 years time.

Baking method heats whole card which isn't something I wouldn't say worth it. ie. G80 die is the problem and it's going to die 100% again.

I would love a hot bed and proper reball setup ..having watched Louis Rossmann and his crew do reballing its not for the faint hearted or anyone with unsteady hands 🤣

So for me if I hit anything that needs that kind of work and its a rare part then its likely to go back on eBay with the issue listed and for repair only, there are a lot of collectors out there that will buy rare parts for repair.

G80 and other Nvidia chips die due to different reasons, which are not applicable to VRAM packaging.

The Serpent Rider wrote on 2023-03-15, 11:02:

G80 and other Nvidia chips die due to different reasons, which are not applicable to VRAM packaging.

Well yes but we were talking more about reballing/reflowing ...one works the other is a temporary method of fixing the problem, replacing Vram BGA does require reballing skills and equipment.

TrashPanda wrote on 2023-03-15, 11:08:

The Serpent Rider wrote on 2023-03-15, 11:02:

G80 and other Nvidia chips die due to different reasons, which are not applicable to VRAM packaging.

Well yes but we were talking more about reballing/reflowing ...one works the other is a temporary method of fixing the problem, replacing Vram BGA does require reballing skills and equipment.

Reballing is most of the times just as temporary as reflowing. The issue is not at the place these solutions target.
https://invidious.baczek.me/watch?v=1AcEt073Uds

RandomStranger wrote on 2023-03-15, 11:19:

TrashPanda wrote on 2023-03-15, 11:08:

The Serpent Rider wrote on 2023-03-15, 11:02:

G80 and other Nvidia chips die due to different reasons, which are not applicable to VRAM packaging.

Well yes but we were talking more about reballing/reflowing ...one works the other is a temporary method of fixing the problem, replacing Vram BGA does require reballing skills and equipment.

Reballing is most of the times just as temporary as reflowing. The issue is not at the place these solutions target.
https://invidious.baczek.me/watch?v=1AcEt073Uds

Oh yea, reballing GPUs is a hit and miss game and you can spend hours hunting down problems involved but Vram usually has a higher success rate. Ive watched Louis for years now and seen a lot of his earlier reballing stuff and the frustration involved with it.

TheAbandonwareGuy wrote on 2023-03-15, 00:13:

If a cards artifacting at BIOs underclocking the VRAM isn't going to help, as its not even running at full power state until you load drivers anyways.

Does the 9700 have more than one clock state? I don't think it has any power management. I think R420 was the first high-end chip to implement it.

swaaye wrote on 2023-03-15, 18:21:

TheAbandonwareGuy wrote on 2023-03-15, 00:13:

If a cards artifacting at BIOs underclocking the VRAM isn't going to help, as its not even running at full power state until you load drivers anyways.

Does the 9700 have more than one clock state? I don't think it has any power management. I think R420 was the first high-end chip to implement it.

rabit.png

This is possible. I plan to do some more detailed troubleshooting on my AIW 9700 later, I'll see if I can confirm.

RandomStranger wrote on 2023-03-15, 11:19:

TrashPanda wrote on 2023-03-15, 11:08:

The Serpent Rider wrote on 2023-03-15, 11:02:

G80 and other Nvidia chips die due to different reasons, which are not applicable to VRAM packaging.

Well yes but we were talking more about reballing/reflowing ...one works the other is a temporary method of fixing the problem, replacing Vram BGA does require reballing skills and equipment.

Reballing is most of the times just as temporary as reflowing. The issue is not at the place these solutions target.
https://invidious.baczek.me/watch?v=1AcEt073Uds

This is true in certain time periods only I think. Note that the video is from 2014 and that's going to be covering bad hardware from ~2008 to 2012 where customers are sending things in for repair. That was the era in which nVidia had their bumpgate, where many of the GPUs and chipsets from that time period failed because of a failure of the balls connecting the flip chips to the PCB-fitting package. I had more than a couple of Sony laptops that were affected and my sister's Macbook Pro 15" 2012 with an nvidia GPU had a couple of repairs until it was retired.

However, In 2003 we were still using leaded solder so the problems of bumpgate and early ROHS solder hadn't happened yet (* all my dead IBM thinkpads with 1st gen lead-free solder though...). A lot of the problems with the R300 are going to be from heavy use, power supply failure and just RAM failure from how hard those BGA DDR memory chips were driven.

Really what we need is something like the modern nVidia tools (MATS/MODS) for detecting RAM faults so that the correct RAM chips can be replaced. Who knows though, maybe it's the memory controller parts of the R300 dies at fault after a lifetime of use? It lasted through the warranty period without issue so no fault there 😉

Samsung 2.8ns BGA chips get burning hot on 9700 Pro, cooling them is beneficial. I think it's cheaper to find new card then get it repaired by a shop, it's not Voodoo 5500. Lowering the clock could help.
9700 non-pro / TX have slower but much cooler memory chips (usually Infineon), and 256-bit bus is wide enough to feed the GPU. Just overclock the GPU.

I've already mentioned that on other threads, but from my anecdotal experience - early Samsung BGA had high failure rate. Killed most of FX5800 cards too, because they were exclusively GDDR2 Samsung BGA. Probably has something to do with high operating temperature and some internal flaws in manufacturing (usual combo).

So I got the R9700 Pro back out, and yes it is a single power level setup.

I tried underclocking, putting a potato chip bag clip on the memory modules to add some pressure to them, etc,

None of it helped. I also noticed when pulling one of those clips off, mannn do those memory chips get hot. No wonder they fail.

Also of note is that atleast mine didn't have thermal paste on it, it had thermal adhesive. That's noteworthy because that could cause different parts of the GPU package to be under different level of tension, making BGA failures that much more likely.

That thermal adhesive is pure evil 😉. That's the first thing I deal with when refurbing R300 card.

stef80 wrote on 2023-03-17, 09:13:

That thermal adhesive is pure evil 😉. That's the first thing I deal with when refurbing R300 card.

Freeze spray or 30 mins in the freezer fixes the adhesive if its epoxy based, used that method on a few cards now !

On NOS cards it goes off easily. On used cards it's 1+ hour job with chemicals, heat, toothpicks and razor blade.

stef80 wrote on 2023-03-17, 10:19:

On NOS cards it goes off easily. On used cards it's 1+ hour job with chemicals, heat, toothpicks and razor blade.

Never really had that issue even with used cards, no .. Ive had one card be a bitch about it and it was a Geforce 4 Ti that looked like it had been ran too long at a high temp.

9700 typically has a phase change material as a TIM. It's solid like a wax at room temp but liquifies with some heat. It is better at gap filling than a thermal paste and the R300 cards have a shim that causes a substantial gap. Thermal paste will eventually pump out of a gap with repeated heating cycles.

The shim should be removed if you want to use regular paste.

Removing the shim is not for the faint of heart. NOS Zalman coolers (VF700 / VF900 / FATAL1TY FC-ZV9) are much safer option IMHO. Base of the cooler is within the shim.