Gigabyte GTX780 OC repair

[MarnickV]

Kind readers,

As you may have noticed I'm new to this forum and to repairing graphics cards in general. I've watched a ton of footage on the interwebs though, so I at least know my way around various components on the PCB, what they do or don't do and what they should or shouldn't do.

That said I have here a broken GTX 780 from Gigabyte. It's the OC Windforce model and I purchased this online off somebody who cheaply advertised this as a dead unit. I figured why not ship out a few bucks, maybe get lucky and have a quick and easy fix. Otherwise have some fun trying to actually repair it.

He advertised it as fans spinning 100% and no signal at any time. Upon arrival I confirmed this and as most would do took off the cooler to do some further investigating. Went ahead and installed the unit in my test rig and wouldn't you know, it fires right up. At that point the first thing that went through my head was BIOS problem, so I went ahead and mounted the cooler back on (without actually plugging it in) and flashed the oldest BIOS suited for the card I could find on it. That however made no difference and during measurement of voltages on the VRM whilst on, my image went and it hasn't returned to date. The fan also doesn't do anything anymore. Instead of spinning 100% like it used to it's now completely dead. A separate cooler yields the same results, but that's hardly surprising at this point.

So I went ahead and verified some things.

The card is equipped with an 8-phase power design plus 2 for the memory.

8 times ON Semiconductor FDMF6823A for the GPU
2 times ON Semiconductor NTMFD4901NF for the Memory

An NCP4208 as main driver for the GPU
And what seems to be labeled as OT=FF D29 as main unit for the Memory (I can't however find any information on this chip but I doubt it's faulty anyhow)

What I've found so far:

Upon startup the GPU gets sent 0.8v for a split second before dropping down to 0.08v and staying there.
There's 12v, 5v and 3.3v present at various points on the card.
The DrMOS units seem fine as they all read 50 Ohm across the VSWH and PGND (not sure if this is enough measuring) whilst off and around 6 Ohm whilst off.
The Main driver also seems ok as the unit measures 5.13v on the VIN, the enable pin is pulled up (~2.3v) and 3.3V is present. The PWM values also correspond with the PWM values on the pins of the DrMOS units. Return line reads the same as the vCore (0.08v).
The fanheader is janky as hell. It reads 12v when nothing is connected but for some reason this drops to 0.9v when any fan is connected??

Prior to the BIOS flash I noticed some purple artifacts (not ever a good sign) in the boot screen but could this perhaps be caused by a power delivery issue somewhere else? Like a part extremely close to failing giving off odd voltages before succumbing permanently?

I would very much appreciate any help with this issue. It's just a little personal project so there's no hurry or anything but I'd love to get to the bottom of this anyway!

Kind regards,
MarnickV

[Rasz]

did you short anything while measuring?
card coming back to life for a short while after taking off cooler might be a symptom of baked GPU

[MarnickV]

did you short anything while measuring?
card coming back to life for a short while after taking off cooler might be a symptom of baked GPU

I don’t think I shorted anything, least I’m not expecting I did. Any telltale signs of a baked GPU? Perhaps a way to measure a short on the GPU itself?

[brian27]

Hi,

I am stumbling upon 'probably' a similar situation here:
https://www.eevblog.com/forum/repair/nvidia-gtx-titan-black-anything-more-to-check-before-i-consider-baking-my-card/

It might be that several enable logic does not work in my card upon some FETs next to the RAM chips.

My question is, I cannot get my card detected by NVFlash, do you by chance get your card detected still after it is completely dead?

[MarnickV]

I'm honestly having a hard time breaking it at this point.
It keeps booting fine right now, just that the fanheader does absolutely nothing anymore.

Upon plugging in the (right hand side) 8 pin PCI-e power connector I heard the sound of electricity flowing. Like when you plug an appliance into a wall socket. I've noticed it only does this when the card will actually subsequently turn on.

//Edit:

The card will turn off after about 20 ish seconds. The vCore drops to about 0.06v. At boot it's about 0.56v. The cooler is entirely off so the chip heats up pretty quickly. The voltage to the fan header is ~9v when nothing is connected and 0.75v when something is. The 3.3v on the same header stays constant no issue.

Not quite sure what to think here. Card turning off after 20 ish seconds to me looks like a possible solder issue under the GPU itself and something a reball might fix. The fanheader issue however I'm completely baffled by..

[MarnickV]

Hi,

I am stumbling upon 'probably' a similar situation here:
https://www.eevblog.com/forum/repair/nvidia-gtx-titan-black-anything-more-to-check-before-i-consider-baking-my-card/

It might be that several enable logic does not work in my card upon some FETs next to the RAM chips.

My question is, I cannot get my card detected by NVFlash, do you by chance get your card detected still after it is completely dead?

As it stands currently the system won't boot with the card plugged in but the 8 pin power connectors disconnected. That to me sounds like the card is still being detected by the motherboard fine?

[brian27]

The card will turn off after about 20 ish seconds. The vCore drops to about 0.06v. At boot it's about 0.56v. The cooler is entirely off so the chip heats up pretty quickly. The voltage to the fan header is ~9v when nothing is connected and 0.75v when something is. The 3.3v on the same header stays constant no issue.

You might be able to try to attach the fan to 12V source just to eliminate overheating issue. But as long as no load applied to the GPU, even without fan, the heatsink itself should be sufficient to cool it down to reasonable temperature.

As it stands currently the system won't boot with the card plugged in but the 8 pin power connectors disconnected. That to me sounds like the card is still being detected by the motherboard fine?

Good then, you can still play around with the BIOS. I am wondering myself whether if the GPU is dead, can the BIOS still be read by the system?
May be some experienced guy can tell me how does the system BIOS communicate with video BIOS during booting? Does the GPU itself involved in the communication?

[Rasz]

You might be able to try to attach the fan to 12V source just to eliminate overheating issue. But as long as no load applied to the GPU, even without fan, the heatsink itself should be sufficient to cool it down to reasonable temperature.

wont do much if the watchdog is shutting the card down

Good then, you can still play around with the BIOS. I am wondering myself whether if the GPU is dead, can the BIOS still be read by the system?
May be some experienced guy can tell me how does the system BIOS communicate with video BIOS during booting? Does the GPU itself involved in the communication?

yes, you need at least somewhat working gpu to bridge the pcie bus, card not showing up means completely dead or shut down gpu

[MarnickV]

yes, you need at least somewhat working gpu to bridge the pcie bus, card not showing up means completely dead or shut down gpu

I'm noticing a peculiar pattern here. Once I leave the card unhooked for a few hours it comes back to life. Leaving it overnight and trying again a day later upon the first boot I get about a minute of screentime before the GPU shuts down. The second attempt gives me about 30 seconds, third 10, etc. This cycle resets once the card has been off mains power for a while. Could it perhaps be capacitor related at this point?

Also does anybody have any idea what's driving the fanheader on this particular card? The voltage on that thing is all over the place and a different value every time I go and check it. It's been anywhere between 8.5v and 11.8v whilst unplugged and between 0.8v and 1.1v whilst plugged in.