GPU oscillator keeps burning out

[hercegovac]

Any insight why 27Mhz oscillator would be burning out on a GPU? Failure appears to be associated with cards high utilization,  when the christal fails card stops working and the voltage to the crystal fluctuates around 0.4 to 0.6V. Card works fine, after the crystal is replaced.

[DaJMasta]

Is it actually a crystal or is it an oscillator?  Is the oscillator being overdriven (like a 3.3V rated running at 5V)?  My other guess would be mechanical stress, but I think it's unlikely in your typical GPU application.


Is there any recovery mode for it?  Like, if it dies, is it possible for it to restart sometimes?

[james_s]

That's definitely an oscillator. Oddly enough I just fixed an oscilloscope that had an identical style oscillator fail, I'd never seen one go bad before that.

[thinkfat]

That's definitely an oscillator. Oddly enough I just fixed an oscilloscope that had an identical style oscillator fail, I'd never seen one go bad before that.

It's kind of hard to see if that's really an oscillator. Crystals come in the same package and the OP has unfortunately chosen to send a picture in an extremely meager resolution. If it is a crystal, the only way for it to fail would be too much drive power from the actual oscillator circuit, which would mean it needs an additional series resistor.

[james_s]

They do? Hm, interesting, I've never seen a bare crystal in a package like that, good to know.

[DimitriP]

My other guess would be mechanical stress, but I think it's unlikely in your typical GPU application.

Thermal stress is more likely. As to why the crystal and not one of the other hundreds of components on the card...that's another story
I took a look at mouser and the "cheaper" ( $0.40 to $0.60)   crystals have max temp of 50C... the "expensive" types  (gotta pay a whole dollar for one) have max temp of 150C

[james_s]

I think the max temp is more a matter of frequency stability than survival, I mean they're made of quartz, there's a reason they use that stuff to make arc tubes in discharge lamps.

Anyway first step here is determine whether it's a plain crystal or (more likely I think) a complete oscillator. Then if it's the latter, check the supply voltage, this would be on pins 2 and 4, so lower right, top left.

[wraper]

On GPU it's quite unlikely to find oscillator as GPU chips accept 27 MHz crystal. Also there are two thin traces running from it and 2 small capacitors on the right side of the photo. So it's safe to say it's crystal, not oscillator.

[wraper]

What is this GPU BTW? It might be that you are just heating faulty GPU chip nearby while replacing crystal. And by doing so, revive it temporarily as it happens with some troublesome GPUs. Nothing to do with a crystal itself.
Also your DC voltage measurement does not say anything. Crystal may oscillate properly with such voltage present on it's terminals. Not to say attaching multimeter leads usually disturbs oscillation. You need oscilloscope to actually check it.

[hercegovac]

It is GTX 1080. I have to happen few crystals  and I keep replacing them, card does get hot, maybe heat then, better pictures attached

[wraper]

It is GTX 1080. I have to happen few crystals  and I keep replacing them, card does get hot, maybe heat then, better pictures attached

IMHO best case you need to reball GPU due to soldering failure. Worst case GPU is dead. Besides heating could be that you just disturb PCB when taking graphics card apart, and it starts working after that. In any case crystal is red herring.

[hercegovac]

Maybe you are right, but after replacing the crystal and adding a silicon thermal pad and additional fan, card has been  stable for a week now.