Zombie Gpu

[Melamol]

Ok, i have a small Problem, well looking at the Gpu market right now, might be a big problem…
I jused an Rx 470 nitro+ 8Gb wich worked well, until it didn’t. One day the pc just wouldn’t recognize it anymore and fans won’t spin. Thought might be something defekt on the pin. First checkt the two fuses at the 8 pin and pcie fuse. Pcie worked, the other two were dead. Replaced it with similar 10A quick fuses. Before testing it again, I checked if the Mosfets were alright, through testing if Vin wasn’t connected with GND. Everything seemed alright. Tested the GPU, it worked. Few minutes later, PC crashed, same story again. Same fuses were dead and this time the Mosfets seemed to be shorted (Vin to GND). Since I didn’t knew wich one, a I detached all 4 and measured them. The dead one seemed to be open (connected) between gate, source and drain. I left it out, just to test the Card again without the short one and without replacing it. It worked again, for a few minutes. Now the two fuses near the 8 pin connector where alright. This time the Fuse near the pcie connector was dead. Also an other mosfet was shorted (the last one down on the Card). Somebody knows what’s going on? Why seems everything going to short after a while? Oh and before someone asks, yes i tried a different PSU actually a brand new bequiet 1200W. I have no more ideas what to do now?!
Picture of Gpu is below.

Edit: A second thought hit me. Is it possible, that the fuse near the pcie connector fries, because I startet the card without one of the Mosfets? So it tried to draw more power over pcie because the one MOSFET was missing? Also I assume, that the last mosfet on the card, is connected tonthenpcie connector instead of the 8 pin. That’s what I got from measuring them. If one gets shorted all but the last one are short to. But now, that the fuse near the pcie broke, this one MOSFET broke too.

 

[RazorWind]

What tools are you using to install and remove the power stages from the board?

When you say you're seeing a short from "Vin" to ground, do you mean the 12V input? What is the exact resistance value you're seeing?

 

[Melamol]

Hey, I use a and soldering station. Small heat gun so to speak. And a tweezer, a bit Kolophonium and soldering paste.

I mean between the big vin and the pig pGND on the Chip (picture below). I already desoldered all 4 Mosfets. And the resistance value from the supposed to be defekt ones are: between Vin and TGnd = 20Kkohm; between Vin and PGND = 16kohm. The ones from wichbinthink are alright are: between Vin and TGnd = shows infinite; between Vin and PGND = 12Kohm
Also i have been told the. Caps might be damaged. So I measured them and between + and - side, the have 0ohm, Sontheim seem Tonne directly connected. Because of this I removed all of the 10 caps i marked in the picture below. After that I measured the spots on the board and still they seem to be connected, 0ohm between each of them. The caps itself seem to be OK. All show around 820 uF. Without Mosfets and caps I seem to have a short somewhere. I don’t know if it’s supposed to be like this though.

 

[RazorWind]

So you removed all the bulk caps, and the power stages, but you still have a short between the pgnd and Vin pads on the board?

 

[Melamol]

So you removed all the bulk caps, and the power stages, but you still have a short between the pgnd and Vin pads on the board?

No, the pads are fine now. The short is between each cap, or where the caps were. I read 0ohm between the +and - side. I marked it in the second picture.

 

[RazorWind]

And the short really is 0.000 ohms?

 

[Melamol]

No, those are fine. The short is between each cap, or where the caps were. I read 0ohm between the +and - side. I marked it in the second picture.

And the short really is 0.000 ohms?

Well 0.3 to 0.4, but i get the same value when I contact both pins of my multimeter. So i guess it’s in the margin of error.

edit: I see now that the caps below the gpu core show 0 Ohm too. This is pretty strange. It might be, that my multimeter is just not sensitive enough and that it is fine likening is.

 

[Melamol]

Oh and if I remove one fet for the 8 pin, does it automatically draw more power over Pcie, wich could blow the fet there?

 

[RazorWind]

Oh and if I remove one fet for the 8 pin, does it automatically draw more power over Pcie, wich could blow the fet there?

Probably, yes. You need to fix the card properly before you attempt to use it again.

You should have something like 1.5-2 ohms of resistance on the core rail, so if you have less than that, you have a short. The fact that the card works anyway means that you've been fortunate so far, and your failure is not in the GPU die. The bad news is, you've got some other component that's in short - most likely either a capacitor or possibly the PCB. You need to figure out which one it is and replace it. You've eliminated the bulk caps and power stages (but created some new problems for yourself WRT reinstalling them).

Visual inspection is your best option. Look for cracked, burned or damaged parts. Failing that, hook up a bench power supply to just that rail, and supply 1.0V at say, 20 amps. Look for components or parts of the board that get hot. Use your knowledge of electronics to decide if the component in question should be getting hot or not, and then go from there.

 

[Melamol]

What/where exactly is the core rail?
Would it be ok if I connect all 5 ground pins on the 8 pin and the 3 12V pins, then hooking up a usb Charger to it (ground and 5V) or maybe a battery is enough (not sure if the defekt component gets hot enough to visualize it Witzbold Batterie though). Also saw that creating a smoke layer above the pcb with Kolophonium might help. The shorted/ hot component will clean the smoke then.

 

[RazorWind]

What/where exactly is the core rail?
Would it be ok if I connect all 5 ground pins on the 8 pin and the 3 12V pins, then hooking up a usb Charger to it (ground and 5V) or maybe a battery is enough (not sure if the defekt component gets hot enough to visualize it Witzbold Batterie though). Also saw that creating a smoke layer above the pcb with Kolophonium might help. The shorted/ hot component will clean the smoke then.

The core rail is the big voltage rail that powers the GPU itself. The terminals of the chokes might be a good place to hook up the positive lead (by soldering it).

Short answer to your second question is no. You need a bench power supply so that you can control the voltage being applied. You need to apply that voltage to specific locations, and not just the whole board. You can use isopropanol, which will boil or evaporate when in contact with hot components, to visually detect which components are getting hot, if any.

 

[Melamol]

Ahh, that’s a bummer. I don’t have a bench power supply.