Всем привет,

I got this card for repair. It had 3 ohms on memories but I located 4 MLCC capacitors on the back which were making a short circuit (KZ) on the memory power. These capacitors looked like exposed to the moisture. This VRM worked under this almost "KZ" state. I replaced capacitors (no thermal image for this condition, sorry)

The memory resistance went back to 95 ohms but this card has very hot memory chokes and it consumes more power on the memory power phase. Everything looks good, all resistances are good, voltages are good, nothing else shines on the thermal cam, bios is being read and was reflashed. Even 1.8V is perfect ~900 ohms. VCORE is initially ~0.9V and goes to ~0.76V when the post code reaches b2. PEXRST works perfectly.

But the card is looping, going to b2, stays there for 20 seconds and retries to initialize card without restarting a computer. I managed to get picture maybe twice for a few hundreds of attempts and even MATS finished and PASSed but only twice. Now it only goes to b2 and again and again.

What do you think friends? Is the GPU dead?
Almost a hundred ohms on the memory but such power consumption? Memory controller? GPU looks good on the thermal cam on the top, no hot spots.

This is normal.

I think you must use oscilloscope to check all phases PWM signal on GPU and memory at first. Also you need check (with oscilloscope ) all other power.

All impulses on all chokes are perfect on the oscilloscope. They are clean, no inductive over shots, all look perfectly clear. The very first thing I do is to measure signals by scope.

Uhh.. I see very hot region and it`s see not normal - or this PS is bad, or some capacitors have a VERY BIG leak - but only on work (on voltage). Try to desolder most (or all) capacitors on this line PS.

If you use riser to test this card - this card must be in "idle state" and don`t be so hot.

maybe you can try a reball. I had a 770 that got hot, and then code 43. After reball everything was ok.

Leakage on capacitors would be seen as hot places. High ESR converts into heat. As I said it was like a water spill or moisture what ate some capacitors and they started to glow in the thermal cam as crazy. I don't have this image but capacitors were corroded and resistance was 3 ohms. This is the main 1.35V to the memory and memory controller.
It was a short, not overvoltage and this card sometimes wakes up until the next reset, at least mats loads from USB drive and passes.
Sometimes the motherboard gives short beeps from the pc speaker, not error beeps but rather such random very short beeps like initialization would have been aborted.

I expect that there is something wrong with the chip because this abnormal power consumption from that phase. The physic's law says there is conservation of energy. If I don't see anything hot on the PCB so this energy from both chokes goes to the chip and it dissipates there.

This almost 100 ohms on the memory is strange, it must decrease when the GPU is powered and the GPU eats this energy. Definitely it is not a "KZ" under the GPU.

Is there a schematic and boardview for this particular card? I did not find. Maybe similar design?

Another question.. This is Asus Strix 1080 but most 1080s have such circuit. What is the purpose of this circuit? It is memory phase 1.35V shorted to ground by 6 x 15 ohm resistors and a power mosfet. What is this for? It does not connect anywhere and to anything, just to ground like a dummy load. The same is for PEX but less resistors.

This is a "discharge circuit" - for an instant emergency shutdown of the memory voltage, in some cases, it follows the signal to turn on the memory power.

This circuit, moreover, very often misleads beginners - since the Q577 transistor remains OPEN even after all power supplies are turned off, and when checking the memory resistance, the multi-meter will show a short circuit - although it is just an open transistor that remains open for about 10 minutes after the board is turned off.


Leaky capacitors do not always get very hot - with a partial short circuit, they can short-circuit the power in the presence of voltage, practically not heating up.

But more and more points to the failure of the GPU. GP102 are very sensitive and very often fail at the slightest reason.

This is very informative post, thank you I thought it was for starting a VRM when it is not loaded to address a light load and possible startup problems. This 10uF in the gate was also strange as it kept this transistor conducting for a long time. This capacitor can be discharged by a reverse leakage of double diode and body diode of the Q576 and a leakage in general.
Now it is obvious


Yes, ~100 ohms in static condition without power and overcurrent when powered on. This does not smell good.
I have seen glowing single memory many times with a good resistance when cold but it is not the case here. Memories look uniformly warm.

My pleasure



You can desolder all the memory bank and check the GPU with "MATS -e 1" = this key testing memory IN GPU (internal cache). If memory in GPU (cache) is well - you will see PASS w/o ANY memory onboard.

Thank you for the tip but the owner wants to sell it for parts on Ebay to another naive buyer This is mystery when it resurrect from time to time the MATS passes. He will not pay for experiments with 11 vrams anyway. I would have to solder them back after tests, no way
The problem is that this card does not initialize, it goes to the "almost beep" and stays there for 10-20 seconds and initialization starts over.

Initially it looked as an easy repair with some capacitors in "KZ", failure as always very easy to spot. Resistance from 3 ohms went back to normal.

There is another theory that there is corrosion under the GPU or in deeper layers and GPU sits in a software loop trying to do something with memory controller and it consumes extensive energy.
I hate to reball GPU because from a minor fault I can cause total failure just because only heating the chip and because it doesn't help in most cases.

But you can only heat the GPU up to 230C with flux - without removing it, this will remove possible corrosion under it. And it will save you the necessary chip reball.

Or without doing anything, sell this board as it is - there are enough naive buyers

This is not mine board The owner will sell it. I would never sell repaired card because it would fail in a month or a year You can buy a new one, run Furmark in 4k and kill such card almost instantly

Maybe I will reflow it anyway. I have everything for IR soldering.

Sounds great, good idea! Give this GPU a chance

After reflow the card became undetected, amen

This shouldn't have made her work worse - perhaps there were short circuits under the GPU. Need full reball GPU((.

Yes, I know but it may be also under memories or even between layers. No time for this card. I think it is corrosion under the GPU and when I remove it I will have missing pads. It's a Rabbit hole. I can't afford more time.

Yes, it makes sense. Good luck with your next repairs!

Hehe thanks, they are ongoing
I had a batch of cards from a customer who purchased some palettes from the insurance company or something. I tested 30 and he gave up. All had PCI-e signals shoted "KZ" or open, refclocks several lanes, 220n capacitors with 20-30 ohms Not one not two capacicors were like a resistors but lets say 10-20. Like after a thunderstorm or taser PEX voltages were killed into KZ. A magic force killed all of them from the PCI-e, total disaster.