ASUS R9 290 vertical stripes

[sonicgear]

First of all hello everyone! I have been lurking on this forum for a while reading GPU repair threads. Now I finally registered to seek some help from you repair gurus.

So my friend asked me to try and diagnose his ASUS R9 290. It is showing static vertical stripes which have different positions at different resolutions.




Also the card will not boot into windows with its' driver installed. The card also had one blown capacitor which I replaced.
We have already tried flashing the stock BIOS but it does not solve the problem.
Checked the voltages and all of them seem to be present. I also measured some of the resistances on the card.



The one thing I don't understand is why there is only 120 ohms between +12V and GND of the 8-Pin PCI connector. The 6-pin connector is showing much bigger resistance.
I suspect that a memory chip might be faulty, replacing it would not be that hard, only problem is how do I find out which one it is.
I would appreciate any kind of help.

[Armadillo]

Firstly, I think you should clean the pcb connecting copper stripes, it looks tainted.

Cause I don't think is the memory chip, as it will affect everywhere.

The only likely causes are from the copper stripes leading up to the CPU output pins.

[sonicgear]

Okay so I just cleaned the copper PCI contacts with 99% isopropyl alcohol but nothing changed.
Most likely RAM failure, the problem is just finding the faulty chip 
http://forums.aria.co.uk/forum/hardware-related/graphics-cards/44703-graphics-card-failure-repair-guide

[Armadillo]

The key word is "Random", not fixed lines..... I don't think memory can fail so coincidentally in lines like that.

I think there is a phrase people use....."Rehot CPU, Bro!"...

I have no experience in that to be honest. But I highly suspect is from the CPU not making good contacts to the output lines. Just my opinion.

Let's hear from others.

[wasyoungonce]

These hot rod video cards really run right on their power hungry limits.  It’s not uncommon to see them fail but 1st have you tried a different monitor?

If it’s blown a cap more than likely the rail it was on has taken a hit.  That PCB looks to have components inbedded in its substrate near the rear connectors and indeed it’s a multilayer PCB. 

In many cases like this the PCB around the CPU is reflowed as they use the an odd non lead solder more akin to metal glue which cracks easily. 

Your only hope is to follow the line where the cap was damaged look and measure the components off that rail but you really need a cct diagram, which would never happen. 

Sorry if it’s not the monitor it’s probably toast.


Sent from my iPhone using Tapatalk

[Rasz]

Have you tried looking if all ram chips get warm at the same rate?
you will probably need pcie extender to run card flat on the table and tall radiator for gpu

The key word is "Random", not fixed lines..... I don't think memory can fail so coincidentally in lines like that.

nah, its classic bad ram

I think there is a phrase people use....."Rehot CPU, Bro!"...

confused, are you advocating it? 

[Armadillo]

Nah!, I did have the last 2 lines of qualification; 

But come to think of it, the Rams are used as image layers, within that bad ram, bad row or column addressing.... yes sounds logically.

OK, I'll go with the Classic truth.

Have you tried looking if all ram chips get warm at the same rate?
you will probably need pcie extender to run card flat on the table and tall radiator for gpu

The key word is "Random", not fixed lines..... I don't think memory can fail so coincidentally in lines like that.

nah, its classic bad ram

I think there is a phrase people use....."Rehot CPU, Bro!"...

confused, are you advocating it? 

[sonicgear]

I found an old chipset heatsink that would do the job, running the card for a minute or two works ok. I also attached a thermocouple so I would know when to shut off.
So I tried feeling the ram with my fingers but all of them seemed to get pretty warm. So i tried putting a drop of isopropyl on each one of them and looking.



All of them seemed to evaporate at about the same rate. The only thing I could tell is that the part circled in purple runs a little cooler than the rest.

Btw you are saying that the low resistance on 8-pin rail is ok?
Here is an image of the backside of the board.



Is there any method the measure the resistance on some of the capacitors under the ram chips and tell which chip is faulty?

[Bashstreet]

Sorry for your card

Best of luck with the repair.

[Rasz]

You can try unerclocking both vram and gpu with bios mod https://github.com/OneB1t/HawaiiBiosReader
have you tried pushing individual ram chips with your finger/clamp and then starting up computer? I found few bga defects using this method

You measured resistances, did you also measure voltages? Terabit Lab and CORE have good videos showing how to measure every rail on AMD cards
https://www.youtube.com/channel/UCRRTfkRDKqsI9FSXP1sc4kw/videos?disable_polymer=1
https://www.youtube.com/user/1servicecore/videos

in the end it might be blown gpu memory controller too only way to tell would be swapping ram from donor

[sonicgear]

Just tried pressing all of the chips really hard several times, nothing changed.
What would happen if i underclocked the card? It seems really complicated to do though.

I also measured some voltages:



Also some images from GPUZ

[Rasz]

Just tried pressing all of the chips really hard several times, nothing changed.

sorry, I wasnt clear, I meant start/test while keeping pressure on individual chips

What would happen if i underclocked the card? It seems really complicated to do though.

if fault goes away you will have confirmation its one of the ram chips, or bad GPU, as opposed to bad connection between
btw stickers on the back look a little baked, maybe someone already tried stone age methods on this card

[fzabkar]

What are the markings on the Elpida RAMs?

I'm wondering whether one could map the RAM layout by shorting certain pins, eg chip selects.

[sonicgear]

What are the markings on the Elpida RAMs?

I'm wondering whether one could map the RAM layout by shorting certain pins, eg chip selects.

Sorry for the late reply, I tried a few more things with the card.
The chips are Elpida W2032BBBG-6A-F, there are 16 of those.

I watched this video of the russian guy diagnosing and repairing a GTX 1080. He has the issue of the card not taking drivers.
https://youtu.be/onoZtI0S9y0?t=164
He tested the card's VRAM with some software and it showed to him which memory banks were faulty. Unfortunately that tool only works on nvidia cards.

I found a similar piece of software and tested the VRAM for errors.



0 errors found... Note that the total memory size seems off, I dont think it reads 4GB.
I am so stuck in a dead end.

sorry, I wasnt clear, I meant start/test while keeping pressure on individual chips

I also tried applying hard pressure on each of the chips while the card was running and connected to a raiser and monitor, the image didn't change at any moment of time.

[wraper]

0 errors found... Note that the total memory size seems off, I dont think it reads 4GB.
I am so stuck in a dead end.

You won't be able to test it without proper graphics driver. Total memory is less than 33MB.

[sonicgear]

Also when I install the driver and launch widnows via my integrated gpu, this software does not give the option to test the AMD GPU. I also tried OCCT, Evga oc scanner X and Gpu mem test, nothing works

[wraper]

Those are just stress tests. Test on the video was not even done by repair guy, he just opened text file with results. He gave GPU to someone who had some unknown professional software to test at low level.

[sonicgear]

You are probably right, that tool gave some very high level detail about the memory.

[wraper]

I'm not 'probably' right. I speak Russian and guy said just that.

[sonicgear]

Oh sorry about that, I didn't realize.
What do you think, can the card be saved?

[Rasz]

You start to sound desperate maybe try reflowing just the ram chips

[fzabkar]

There are 10 signal traces between each RAM and the GPU on the top side of the PCB. If you are willing to take a risk, try shorting each of these traces to ground, one by one, until you can map the associated RAM to a corresponding part of the screen. Each trace has a pad designed for this purpose. :-))

[sonicgear]

There are 10 signal traces between each RAM and the GPU on the top side of the PCB. If you are willing to take a risk, try shorting each of these traces to ground, one by one, until you can map the associated RAM to a corresponding part of the screen. Each trace has a pad designed for this purpose. :-))

WOW! You are a genius, following your advice i located which chip doesn't distort the picture when shorted to ground, thats the one that died!
There are actually 12 vias for each chip but only shorting one did the job. Thank you so much!
I will be ordering a replacement chip, a gddr5 stencil and some bga balls. I read that 0.4mm balls for gddr5 are okay?
Will be updating the situtauion once the stuff arrives from aliexpress.

Thank you very much once again!

[wraper]

I will be ordering a replacement chip, a gddr5 stencil and some bga balls. I read that 0.4mm balls for gddr5 are okay?

If you buy a new chip, you don't need to reball it.

[sonicgear]

Okay, thank you. I will report back with my results