AMD gpu black screen or vertical stripes problem cause?

[mikkh]

So as most of you there is grey lines or black screen crash apparent (on some of the GPU´s) since HD5000 series and even up to R9 2XX (maybe even newer cards too), but what causes it? Most of the time when the GPU core and memory are underclocked the crashes seem to disappear for a while and the grey screen crash becomes black screen crash.
Could the cause be electromigration? But doesn´t seem to because the crashes sometimes happen on less than a month old GPU´s.
More into the problem, I have seen newer nVidia (for example GTX780) GPU´s crash the same way.

The crashes seem quite random and don´t look like they are caused by bad power supply part of the card.
What is more, sometimes some of the faulty black screening AMD GPU´s work excellent in secondary crossfire.

Has anybody had some experience in repairing them and what could be the solution?

[SaabFAN]

Have you checked your voltages on the PCIexpress-Supply connectors? How much Watts can your Power Supply deliver? Which manufacturer made your Power Supply?

What you describe sounds like the Power Supply is unable to deliver enough juice to the card and once it has to do something the voltages drop too low to distinguish between 1 and 0 on the signal-lines.

Once you're sure the PSU is not the problem, you can start looking at the Graphics-Card itself. If you have an Oscilloscope, you can check for ripple-voltage on supply-rails of the card itself and if they look okay, then you are probably dealing with a faulty chip or a cracked solder joint on the Graphics Card.
Also, if possible, check if the cards work in another PC. If they do and you eliminated the PSU as the cause of the fault, you are likely to look at a problem on the Mainboard.

[SAUL BRITTO]

I agree with SaabFAN and more: Memory,generally causes this defect.
 

[mikkh]

Well I am thinking of trying to find old griddle for preheating the board and adjustable temperature heatgun for the top part of the card.

Sometimes the AMD gpus have horizontal stripes with drivers installed, but these stripes won´t appear in bios etc and without drivers. Seems like a bad gpu solder joint to me?

[wraper]

Seems like a bad gpu solder joint to me?

Yet another one 

[wraper]

On particularly R9 290/290x you can expect high probability of problems with RAM ICs themselves. Sometimes can be solved by using better cooler to keep RAM chips cold.

[System Error Message]

the black screen crash is a driver bug in relation to power savings (PC idles, monitor goes to standby, GPU goes to standby and cant wake up).

Otherwise the issue is usually with memory. Do a stress test to test the GPU core and memory seperately (there is even memtest for GPUs).

[Rasz]

There is a reason to stick with top brands like ... Dataland

WAT? Tul is a bottom of the barrel discount manufacturer, Powercolor, VTX, dataland (lol) are all supermarket brands.

from my very limited experience(couple of cards) ATI/AMD cards usually suffer ram/power supply fail
on the other hand there is something like 1:100 if not 1:10000 defect rate ratio between amd and nvidia. Basically EVERY SINGLE nvidia card up to at least FERMI will die from faulty(by design, planned obsolescence?) GPU sooner or later, and usually its sooner.

[mikkh]

Seems like a bad gpu solder joint to me?

Yet another one 

The small solder dots on the back of the card seem to be oxidized and there are tiny "tin whiskers" .

But crashing with BSOD and then artifacts with drivers installed seems to be somekind of solder issue? The PCB is a bit bent too. I remember having older bent AMD card (black screen on startup) and I just took the cooler off and left it on  the table for a few weeks  and the card works ok still.

[wraper]

There is a reason to stick with top brands like ... Dataland

WAT? Tul is a bottom of the barrel discount manufacturer, Powercolor, VTX, dataland (lol) are all supermarket brands.

Tul is perceived as budget for some reason. But IMO this is more rumors than how things actually are. There usually is not difference in reliability if you buy reference design. Most likely they will be the same cards from the same factory with different stickers on the cooler. Especially if graphic card was launched recently.  What you might care is how warranty is provided in your location. But you know what, if you are in EU the store where you bought it no longer exists, you can still deal with them directly. Though you will pay for the shipping both ways. For example with XFX you have no warranty in that case, but they offer good service in US on the other hand.

[wraper]

The small solder dots on the back of the card seem to be oxidized and there are tiny "tin whiskers" .

I rather suspect this is a dust  You can try to heat the chip to say 130-150^o C, if it starts to work normally after that, then you can be sure that it is faulty.

[mikkh]

Nah that is not dust, I cleaned the GPU with 99% IPA. But if the heated (220C) chip lasts 1-2 years I would be more than happy. So the problem is the faulty underlayment on flip chip BGA´s? Why don´t the chinese reball those flip chips then? That way the chips would last a lot more.

[wraper]

Nah that is not dust, I cleaned the GPU with 99% IPA.

With IPA will make a swamp of dirt unless submerge the board completely and use ultrasonic cleaner.

But if the heated (220C) chip lasts 1-2 years I would be more than happy.

Extremely rare and only possible if not heavily used, expect a few weeks.

Why don´t the chinese reball those flip chips then?

If you succeed doing this cheaper than GPU cost itself, then I'm sure you will qualify for million dollar prize.

[mikkh]

So what is the cause then - electromigration or the underlayment of the chip problems? Itf the chips are kept at constant temeprature all the time they should last until electromigration?

I used to fix through hole mount old car dashboards for friends by just redoing every solder joint on the pcb with a soldering iron. Well the repairs only lasted 10 years, because I didn´t use flux on them.