GTX 1080: story of the fallen MOSFET hero

[wraper]

Want to tell about some unusual graphics card repair case. A while ago I bought a faulty EVGA GTX 1080 SC (reference PCB). It had shorted 12V power. Turned out to be shorted dual MOSFET in RAM VRM, thankfully nor RAM, nor GPU died due to overvoltage. Replaced faulty MOSFET, checked VRM with oscilloscope, all worked fine. Assembled it back, run some furmark and Unigine Valley tests and it worked fine. Then after a while put it onto test stand, pressed the power button, and PSU had gone into protection. I thought like, fuck, RAM VRM should have died again and RAM and/or GPU are probably dead now  . Already had such precedent with GTX 1080. Disassembled it, and nope, it's shorted Vcore VRM dual MOSFET NTMFD4C85N. Replaced it, and card started working again  . Time to check all Vcore VRM phases. Only one out of 5 phases was active. I'm like: OK, card might run only on one phase due to 2D load. Started Unigine Valley, and fucker still runs on one phase. And runs the test JUST FINE on one power phase!!! Turned power off immediately to not burn the thing again. Then found single/multi phase mode input of PWM controller. GPU properly gives the signal to switch into full phase mode when 3D load is present. Strange... Checked control outputs from PWM controller to MOSFET drivers, all 5 phase signals are present, even stranger. Turned out that 4 out of 5 MOSFET drivers had no power. On 12V power input they have 2.2 ohm resistors and MLCC decoupling capacitor after that. On all 5 phases caps were either shorted or had resistance in tens of ohms range. 4 out of five resistors failed open. And only one phase kept working (which BTW is also the phase which is active in 1 phase economy mode).
Apparently capacitors kept failing short until only one phase was left alive. However this tenacious card still did not give up against all odds, until another unrelated RAM VRM failure happened. And this poor heroic MOSFET only gave up after I repaired initial failure and run some stress tests  . And did not even die during those stress tests but only after that on power ON. Also strange that uP9511 PWM controller did not go into protection. It has current limit and shutdown functionality according to the datasheet. As of why capacitors failed, I can only think of manufacturing error with low voltage MLCC mounted instead of proper part.

[Runco990]

Dang.... what a saga! 

I hug my 1080ti ftw3... it was the last card I could buy at msrp.  Still running strong!

To hell with the scalpers.

[Kjelt]

That is some perseverance you showed there.
How do you measure those cards under load, all those huge heatsinks in the way and all, do you have some kind of extender board and alternative cooling setup ?

[exe]

Wow, I'm always impressed by repairs of videocards and motherboards. To me looks almost like magic.

[Shock]

Cascading failures are the worst time sink ever. So I take that it's running ok for now? You should do a full verbose brain dump of your complete troubleshooting process at every turn. It would make for very interesting reading kind of like how Shahriar does it on "The Signal Path" videos.

[wraper]

Cascading failures are the worst time sink ever. So I take that it's running ok for now?

Yes, it's working. If MOSFET drivers were not in tiny DFN package, I would find the problem much faster. Measuring something with 0.5 mm pin pitch, and what can destroy the whole card if you look at it funny, is not fun when card is powered.