[Update: It LIVES] Fixing a Dead PC Graphics Card

[iamdarkyoshi]

I bought a 41$ r9 270x on ebay to repair. I was hoping for the usual shorted mosfet, but unfortunately... no. All VRMs work fine. The card is recognized by the PC, (using it as a secondary card) but as soon as the card does something (like running folding@home or a monitor is connected) the PC hangs. I have already sucessfully reflashed the card's BIOS. All I can think of is that it has defective videoram, faulty GPU die, or something needs to be reflowed. The listing said that the card was never overclocked or abused, it just stopped working one day.

Where should I start?

[Tim F]

Does it hang immediately as you plug in the monitor or run for a bit before it hangs?

If it works for a bit before it hangs, try having the card do something that wouldn't generate enough heat to make it hang, then heat it up with an external heat source.

[iamdarkyoshi]

Does it hang immediately as you plug in the monitor or run for a bit before it hangs?

If it works for a bit before it hangs, try having the card do something that wouldn't generate enough heat to make it hang, then heat it up with an external heat source.

literally as soon as a monitor is connected. Cold boot on the PC with a monitor on this card results in a complete crash, dorsnt even have to be booted into an OS.

[Rasz]

what about in dos?
put oscilloscope on vgpu and vram and look at the rails when it crashes
you could try swapping ram

[iamdarkyoshi]

what about in dos?
put oscilloscope on vgpu and vram and look at the rails when it crashes
you could try swapping ram

Thats a thought, maybe the VRMS cannot provide enough power to actually run correctly, just enough to let the card boot up. Swapping videoram will certainly be fun. That might be a last resort...

[kripton2035]

reflow the GPU (just reflow, no need for desoldier and resolder the chip) to see if symptoms change.

[iamdarkyoshi]

reflow the GPU (just reflow, no need for desoldier and resolder the chip) to see if symptoms change.

I am just wondering if a reflow is what is needed here. Its a pretty recent card...

[SaabFAN]

From my experience, a problem caused by the BGA-Package that can be solved by reflowing the chip manifests itself rather differently: Artifacts once the card gets warm, sudden crashes with otherwise normal operation, etc.
The problem you are describing sounds more like the symptom of a defective GPU-Chip.
This could be caused by a rather nasty effect called electromigration: Atoms in the coducting elements are pushed out of position by the electrons moving through the conductors and pile up in other locations where they cause a short-circuit. Normally, Chips are designed in a way that this effect doesn't destroy them too quickly, but when operated at high temperatures, or applying excessive voltage to reach the desired clock-rates (which can happen even without the user playing with overclocking just because the manufacturer decided to increase the voltage until the chip works instead of tossing a bad chip), then it is entirely possible to see a GPU die within one or two years.
The Pentium 4 Northwood-Cores were particularly vulnerable to this effect, leading to the "Sudden Northwood Death Syndrome".

Unless there are some other hidden problems in your card, which I doubt  (basically everything except for the RAM and the Power-Supply is integrated into the GPU today), there is no way to repair it.

[rollatorwieltje]

Have you checked the clock speeds and other overclock settings? Maybe someone tried to overclock it and stored some bad values in it.

[iamdarkyoshi]

Have you checked the clock speeds and other overclock settings? Maybe someone tried to overclock it and stored some bad values in it.

Crashes even before POSTing and I have reflashed the card's BIOS.

[iamdarkyoshi]

It works! I used my hot air reflow station on the GPU and VRAM and lo and behold, it WORKS! I have been running games on it for a while, no thermal concerns and it has been completely stable, no artifacting or anything.

[kripton2035]

told you ...

[iamdarkyoshi]

told you ...

Ya

Now that I think about it, with everyything else working, a bad soldering joint was probably the only repaining issue. bought for 41 bucks, gonna sell for 3x that! Awesome. Imma buy more XD

[kripton2035]

as Blueskull said, this may only work for some months, you should desoldier and resoldier the bga chip
but that's more difficult.
if the card is for you, then reflow it in some months when it does not work again
if you want to resell it, then a desoldering is necessary

[BradC]

Agree with the other posters. "reflowing" is a nasty hack perpetrated by those looking for a quick and dirty fix and it won't last. You might get lucky, but chances are anywhere from a week to 6 months and it's back in the dead pile. If I bought that card from you as a going concern I'd be pretty damn pissed off when I found out what you did. If you have any conscience at all, don't sell it.

I've done it on an old Mac motherboard to see what the fuss was about. I then bought some stencils and balls and re-balled some HP laserjet Jetdirect 615's with a hot air rework station to see how to do it better (I won't say properly, but it's a hell of a lot better than "reflowing"). The Mac lasted 6 weeks, the 615's have been going for 2 years now.

[orbiter]

If you're going to sell it as 'repaired', and with some sort of warranty against that repair failing then fair enough.

Personally though I wouldn't sell it, as it will likely fail again and sooner rather than later. Concern for the buyer would get the better of me. Think how you'd feel
if you'd bought a GFX card as working, then found out it had been baked due to a fault.

[iamdarkyoshi]

I will have a 1 year repair/replacement policy on the entire PC, as I am selling local.

[Rasz]

agaik there are no gpu problems in amd GCN chips, so I would be leaning towards  bad connection under one of memory chips = simple reflow should be ok as a permanent fix

[iamdarkyoshi]

agaik there are no gpu problems in amd GCN chips, so I would be leaning towards  bad connection under one of memory chips = simple reflow should be ok as a permanent fix

Ya, I havent seen any AMD chips need reflows... I talked to the ebay seller I bought it from and he said he has absolutely awful airflow. The PC I plan to sell with the card has fantastic airflow.

[Tim F]

A reflowed bad solder joint is usually still a bad solder joint. Running the card cooler will just extend the amount of time that it takes for thermal cycling to crack the joint again. Most mid to high end graphics cards have the heatsink entirely shrouded and the fan speed is regulated by the GPU temperature. The result is that unless the card is already running it's fan at 100%, with improved case cooling it will reach about the same GPU temperature under load just with the fan not working as hard.

A reball is the only way to ensure a decent lifespan before the card needs fixing again.

I'm with the others on this - if you sell the card, even if you offer a warranty you should be upfront about what has been done to it.
If it were me I'd only ask for enough to cover the purchase price, time spent on it and fees/postage. Be upfront about the questionable reliability. The buyer deserves to know that they are taking a gamble just like you did when you purchased it non-working.
Either that or just keep it for myself - if it goes another few years then great, if it kicks the bucket next week then no harm done.

[iamdarkyoshi]

A reflowed bad solder joint is usually still a bad solder joint. Running the card cooler will just extend the amount of time that it takes for thermal cycling to crack the joint again. Most mid to high end graphics cards have the heatsink entirely shrouded and the fan speed is regulated by the GPU temperature. The result is that unless the card is already running it's fan at 100%, with improved case cooling it will reach about the same GPU temperature under load just with the fan not working as hard.

A reball is the only way to ensure a decent lifespan before the card needs fixing again.

I'm with the others on this - if you sell the card, even if you offer a warranty you should be upfront about what has been done to it.
If it were me I'd only ask for enough to cover the purchase price, time spent on it and fees/postage. Be upfront about the questionable reliability. The buyer deserves to know that they are taking a gamble just like you did when you purchased it non-working.
Either that or just keep it for myself - if it goes another few years then great, if it kicks the bucket next week then no harm done.

My listings are always truthful and I will provide information as to what the card has been through.

[SaabFAN]

Interesting... From what you described, I was sure that there was some problem within the chip, not with the connection Chip to Board.

[iamdarkyoshi]

Interesting... From what you described, I was sure that there was some problem within the chip, not with the connection Chip to Board.

That is what I would have thought. PERHAPS cleaning off the thermal paste (which was all over the caps next to the silicon die) fixed it? I measured the paste and did not get any continuity though...