Help repairing a graphics card with a short circuit

[ArnauVF4]

Hi, first of all, I'm a bit of a noob with electronics, I know the basics, but i'm far from an expert.

I got this HD7850 that has a short circuit and triggers short circuit protection on the PSU, so it doesn't boot.

The picture with the red markings indicates the pins in the 6 pin connector that are shorting to ground. There are supposed to be only 2 ground pins, as the diagram in the other picture indicates, but that isn't the case. Also, the i've measured the resistance in the shorted pins and it's 0Ω, so total short there.

At this point, I don't know what components and how to test them. Maybe the mosfet are the problem? But, again, I don't know how to test them. I looked up the schematics but I can't understand thr pinout. All of the mosfets are the same, 4935N . There are also R47 inductors, but I don't know how to test them. All I have is a multimeter.

Thanks.

Enviat des del meu Mi MIX 2S usant Tapatalk

[kosomsk]

Hi! do you have an adjustable psu?

[ArnauVF4]

What do you mean by adjustable psu? Lab bench power supply? I mean I have one, but I don't wanna destroy it

Enviat des del meu Mi MIX 2S usant Tapatalk

[bob91343]

My suggestion is to measure the resistance between pins that are not supposed to be shorted.  You say zero Ohms but of course that's impossible so you need a better Ohmmeter.  You can build one easily with a few parts; I did that and can resolve about 0.01 Ohm with it.  I also have some very good 4-wire resistance bridges that can resolve much smaller resistances and they too will expose the offending part.  (I can measure the resistance of a screwdriver shaft, for instance.)

Once you do that you can probe around to see where the lowest resistance is and there will be your short.  One approach is to measure along a circuit trace to see in which direction the short lies.  As your probe approaches the short, the resistance gets lower.

[wraper]

Try measuring MOSFET gate resistance. Shorted MOSFETs usually have shorted gate as well.

[ArnauVF4]

I was doing it wrong. I put beeping mode

0.01 Ω Reading as you said

[goaty]

You can also power it up with a bench supply and use either FLIR thermal camera or IPA and see where it goes hot.

Oh, forgot, and of course the Al-Electrolytics could have a short as well.

[Psi]

First thing I would check is all the FETs

Then i would look at the gate driver IC's

A low voltage fed from a lab PSU, like 2V, can be used to get a little current flow, maybe 1A, and see what fet/chip starts to get warm/hot. That's a good sign of where the fault is.  However this wont work in all cases, but it will catch some issues and is easy to try.

Also have a good look over the board for any components that may have been knocked off the pcb. This is common with cards as people have screwdrivers around them and slip.
Comparing your card to a pic of a good card of the same type can be a good idea.  (google image search for HD7850  pcb  )

[ArnauVF4]

I think I may have found the problem. I literally took a picture in that spot and I somehow didn't notice. There is one capacitor that's broken! I may attempt to remove it when I can and see if the short dissapears. How would I know the value of the cap? 


Enviat des del meu Mi MIX 2S usant Tapatalk

[Psi]

How would I know the value of the cap? 

Look for another cap on the PCB that is the same physical size and color.

Chances are high it's the same value.

In any case, i doubt it matters too much,  extra capacitance is usually fine.
Figure out the footprint size and lookup on digikey the biggest ceramic cap available that will fit that footprint and which looks the same physical size (ignore any super expensive caps).

It's also possible it will run fine without that cap. or maybe just less stability.

[kosomsk]

What do you mean by adjustable psu? Lab bench power supply? I mean I have one, but I don't wanna destroy it

Enviat des del meu Mi MIX 2S usant Tapatalk

You can also power it up with a bench supply and use either FLIR thermal camera or IPA and see where it goes hot.

Oh, forgot, and of course the Al-Electrolytics could have a short as well.

Simple idea for every day 

[ArnauVF4]

Ok, I got the cap out and the good news are that it now doesn't trigger short cirtcuit protection and there is no short. Bad news are it doesn't show an image and the GPU core doesn't heat up (that means no power I suppose).

Maybe the cap is needed because it is just after the power input? I may be able to steal one cap from an old dead board that I have.

[ArnauVF4]

Big update! I put a cap from a scrap board that looked similar and bam. It booted.

https://m.imgur.com/a/xcJoCfX

Sadly it's giving me vertical green lines which indicate that one or more vram chips are faulty. I may try to buy new chips and a stencil and try. The card has 8 memory chips, and one seller on aliexpress (my only feasable option) has a pack of 4 chips. Should I buy 4 or 8 chips? I don't think more than 4 are defective.

[Psi]

If you have never soldered BGA ram chips before you will need some practice.

See if you can find a dead board you can practice on.

Also, check if those aliexpress ram chips have balls or if you will have to reball them yourself.

Might be worth a reflow of those ram chips, just in case the issue is a broken solder ball rather than a dead chip.
The smashed capacitor may indicate the board was dropped which could have cracked some of the balls.

[ArnauVF4]

I'll definitely practise before. I think i'll buy a stencil pack that includes a ddr2 ram stencil, as I have a dead stick somewhere that I can use to get familiar with bga. They include balls supposedly, then I wouldn't need the stencil right? I may end up buying the stencil just for practising and maybe some solder paste?

As for reflowing, what temperature I should set my station to? And how much air (the knob goes from 1-8)?

Another note, the card was card was given to me by a friend, he told me it first showed the vertical lines (randomly, one day it stopped working, no damage) and then he dropped it or something and then the cap broke and made the short.

[wraper]

Sadly it's giving me vertical green lines which indicate that one or more vram chips are faulty.

Not necessarily. It may also be soldering failure under VRAM, GPU, or faulty GPU itself. Considering cost of working HD7850, I would not bother. It's troublesome to repair. Especially if you have no experience. It will likely end up as throwing good money after bad with no positive result, and probably completely dead card. With Nvidia there is at least leaked test program which can indicate on problematic RAM chip, no such for AMD AFAIK.

[ArnauVF4]

Sadly it's giving me vertical green lines which indicate that one or more vram chips are faulty.

Not necessarily. It may also be soldering failure under VRAM, GPU, or faulty GPU itself. Considering cost of working HD7850, I would not bother. It's troublesome to repair. Especially if you have no experience. It will likely end up as throwing good money after bad with no positive result, and probably completely dead card. With Nvidia there is at least leaked test program which can indicate on problematic RAM chip, no such for AMD AFAIK.

I'm 99% sure it's the ram chips. Why? Because even when the GPU is not rendering anything, there are still the vertical lines. That means that when a 2D texture gets to the memory, it gets corrupted and shows it onscreen. Also this is a common problem with old-ish amd cards.

[wraper]

Sadly it's giving me vertical green lines which indicate that one or more vram chips are faulty.

Not necessarily. It may also be soldering failure under VRAM, GPU, or faulty GPU itself. Considering cost of working HD7850, I would not bother. It's troublesome to repair. Especially if you have no experience. It will likely end up as throwing good money after bad with no positive result, and probably completely dead card. With Nvidia there is at least leaked test program which can indicate on problematic RAM chip, no such for AMD AFAIK.

I'm 99% sure it's the ram chips. Why? Because even when the GPU is not rendering anything, there are still the vertical lines. That means that when a 2D texture gets to the memory, it gets corrupted and shows it onscreen. Also this is a common problem with old-ish amd cards.

1 data line to RAM fails, and you have this issue. Fault can be anywhere, in GPU, in RAM chip, or between them.

[Psi]

If the card was doing the lines before it was dropped then it's probably dead ram rather than ram with broken balls.

[wraper]

If the card was doing the lines before it was dropped then it's probably dead ram rather than ram with broken balls.

You don't need to drop it to cause soldering failure. FYI, graphic cards PCBs are almost always under bending stress due to pressure from GPU heatsink and thermal cycling. And as you may notice there are multiple mounting holes near to RAM and GPU around which PCB will deform. In any case, if it was fully dead RAM chip, it would be way worse than single line.
What I'm saying, trying to fix this is pure gambling. And even if you successfully replace all RAM chips, chance to fix it is 50% at best. If you could reball GPU, chance would be better, but it needs more serious equipment than simple hot air. And replacing GPU won't make any economical sense at all. Not that it has any economical reason to fix it anyway (considering difficulty of repair) because you can buy working one for $15-$30.

[ArnauVF4]

Well, as you all said, It's not worth repairing. If it was a 500$ card, maybe, but for a 40€ card, not worth it

[Rasz]

use a finger debugger, power card laying flat plugged to pcie extender, load a test program and push individual ram chips one at a time. I had great luck identifying cracked solderballs under ram chips this way (in graphic cards and TVs)

[Deus]

They include balls supposedly, then I wouldn't need the stencil right? I may end up buying the stencil just for practising and maybe some solder paste?

If you want to place the solder balls one by one manually you indeed don't need the stencil.
If you want to make it easier, you'd use the stencil. It's meant to align the solder balls to the BGA pattern.


Why not to use a heat gun only on large BGA's for reballing or reflowing: