HD 5770 overheats and shutsdown

[lordvader88]

I have a HD 5770 that displays stuff, artifact free, but immediately starts overheating the core. The fan reaches full speed in a minute, and after 2-3 minutes the video cuts out.

It's the GPU chip itself that gets superhot, so does that mean there's an internal short and no way to fix it ? There's test points all over the place, what if I apply 12V to the gpu off a bench supply while not hooked to the PC ? Or should I really just solder up some wired test points and check when it's on the PC

[SilverSolder]

Just thinking out loud -

Could it be a mechanical problem, e.g. the heat sink partly lost contact with the GPU chip?  - maybe re-install the heat sink with new thermal paste?

The other thing that could have happened is a bad temperature sensor that reads too high?

[lordvader88]

No it's not that and it physically is super hot, and it's not the VRAM making that kind of heat.

This is just something to play and learn with. Next year I'm making or getting an IR heating station for stuff like this.

[SilverSolder]

Another idea that could explain the symptoms:   There might be a local voltage regulator on the board that is supposed to suppy the GPU with a reduced voltage...  it could have failed so it supplies too high a voltage to the chip -  that would explain why it still works, yet gets super hot.

Might be easy to check the voltages on the board and see if they are right?  -  maybe there is software that can read it?

Maybe the voltage is adjustable in BIOS, just like for a CPU, and someone set it too high?

[Rasz]

The other thing that could have happened is a bad temperature sensor that reads too high?

a known problem in 78xx GPUs, diode reading temp shorts inside the package, card runs fine but you end up with fans at full blast

[Gribo]

The thermal interface material might be faulty. Usually, the cheap cards use thermal paste that lasts only two or three years.

[Bashstreet]

Do the normal tests.

1.Check the temperature at temperature logging program and inform us of the results.

If temperatures reach high level on idle proceed.

(a) Check fan is working.
(b) All cables including fan header are attached
(c) Card is firmly in slot and there is no slop in the slot or connectors.

If card still heats up proceed.

(a) Check core voltage and possible overclocking with proprietary program provided with your graphic card (inform of these values)
(b) If no overclocking over voltage is set remove the card from slot and do mechanical evaluation for any visible damage or and loose heat-sink(s)

If all this is negative you will need to proceed to next steps depending of your capability and dexterity if you wish to proceed inform of this.

[lordvader88]

No I used new TIM on it right after the 1st time I tried it. It's not anything mechanical.

To me, it actually looks like there's 3 area's of the PCB that are dis-colored. They are right around the GPU die, but closer to the VRAM than the GPU die. Didn't seem to wash off with rubbing alcohol.

When multilayer PCBs get a bit cooked inside, can they get discolored on the surface ?

I really want to measure the current to it, I should wire up an adapter and measure what the PCI-e cables are delivering.

As for testing these out of circuit, I'm wondering what all logic signals are supposed to be present for them to even turn on, when not connected to a mobo. I should try anyways. 1st I need to make sure the fan will work directly off 12V w/o PWM, and then I wonder if the GPU will work if the fan is not detected.

I know chips will die in seconds without a heatsink. When it shuts down, the backplate that holds the heatsink, is almost 2 hot to touch. Something has to be drawing way too much current.

[macboy]

I would definitely suspect too-high core voltage caused by a faulty regulator.  Power increases with the square of voltage increase.