всем привет, Hello everybody.

Sorry I do not speak Russian, forgive me please.

I have faulty 1070 with all good resistances and voltages. Only voltage of memory VRM (uP1666Q) is 1.55V but memory resistance is 109 ohms.
I checked almost everything, PEX reset is present but not buffered by a logic gate but pin does not look damaged. The LANE0 both pairs give 0.48V voltage drop in the diode mode and they also look good. Back capacitors 220n also look good.
I have a thermal cam and I don't see any signs of failure, GPU is evenly becoming warm, not hot, just warm.
The problem is that the GPU is in permanent reset state. GPU does not start reading from the bios chip. Chip select of the bios chip goes high due to the pull up resistor to +3.3V but nothing happens. On the oscilloscope no transients on any pin of the bios chip and the card is not detected. Power to the BIOS is OK.
Maybe there is some kind of additional power good detection on the board or some kind of additional logic which keeps the GPU in the permanent reset state?

Do you guys have a board view and schematic diagram for this particular PCB? Or maybe you have any suggestions?

pex reset voltage drop?

GPU chip can be bad. I had a reference PCB 1060, with the same behavior. I tried reball, but still the same. after replacing the GPU from a burned PCB -> OK

how do you know it's in reset state ? what's the vcore voltage?

Чип GPU может быть плохим. У меня была эталонная печатная плата 1060 с таким же поведением. Пробовал реболл, но все равно. после замены GPU с сгоревшей PCB -> ОК

как узнать, что он в состоянии сброса? какое напряжение vcore?

всем привет,

All voltages go to the listed values and I don't observe any changes.

1. VRAM 1.56V @307kHz ??? Not too high ?? (I don't have replacement chip now uP1666Q)
2. VCORE 0.821V @303kHz
3. Left VCC_5V 5.04V
4. Right VCC_5V 5.08V (There are two five volt sources on the board)
5. VPEX 1.015V
6. V3.3V on the PCIe slot 3.31V
7. PEX_RESET 3.365V !!! ??? it is higher than 3.31V but maybe the motherboard does it like that.
8. V1.8V is 1.836V

When powering the board PEX_RESET goes HIGH then LOW and then back HIGH. 0 ohm resistor bypassing missing AND gate is OK, I measured it there.

I just measured it again 10 minutes ago because I made a reflow of the GPU on the machine. It did not change anything. No changes at all. This reflow did not break or heal it.
VRMs are warm, GPU is warm uniformly warm, nothing is overheating under the thermal cam.
Posted after 18 minutes 2 seconds:


VCORE should go to 0.8V+ and fall back a little to a 0.7V+. Hardware reset to the PEX pin may not be the only one reset of the GPU. Maybe some other constraints must be met before GPU starts executing. Maybe there is some kind of power sequence or power integrity check, internal reset, brown out, overvoltage etc. This 1.56V on the VRAM double VRM uP1666Q intimidates me a little. I measured internal 2.000V reference pin of the uP1666Q and it is indeed 2.0001V. Resistors around looks good, they have finite values, but what exact values should they have? Maybe this +0.2V prevents GPU from starting?
Who has such board please measure resistors around uP1666Q.

можно онанировать левой рукой, можно правой.
можно попробовать голландский штурвал.
но убитый по резету гпу не оживёт от этих манипуляций, всё, его надо выкинуть в ведро.
ну и, экономным вендорам - вечную ска, икоту. сэкономили 5 центов. или сколько там за буфер.

for me this is also true, a dead gpu is a dead gpu.

But there are rare occasions when a dead GPU is measusing good resistance, and it's not seen by bios. When the memory controller dies, it can still give errors in all mats.

yesterday, I replaced three GPU's on a MSI V330 board, and all of them behave the same as described above. maybe all three are dead in the same way.

I have another board. A pallit, that had a 1070 on it, but the fans small PCB decided to burn.... I restored the board, it gives voltages. I installed an 1070ti on it, problem is that mats gives errors on all banks. I will take the gpu down and put one of the 1070 from the MSI V330 board. If it behaves the same, and it's not seen in bios then it's not something with the MSI board, it's the GPU.


для меня это тоже правда, мертвый gpu - это мертвый gpu.

Но бывают редкие случаи, когда неисправный графический процессор показывает хорошее сопротивление, а BIOS этого не видит. Когда умирает контроллер памяти, он еще может выдавать ошибки во всех матах.

вчера я заменил три графических процессора на плате MSI V330, и все они ведут себя так же, как описано выше. может быть, все трое мертвы одинаково.

У меня есть другая доска. Паллит, на котором было 1070, но небольшая плата вентиляторов решила сжечь .... Я восстановил плату, она дает напряжения. Установил на него 1070ti, проблема в том, что коврики выдают ошибки на всех банках. Я сниму gpu и поставлю один из 1070 с платы MSI V330. Если он ведет себя так же, и этого не видно в BIOS, значит, дело не в плате MSI, а в графическом процессоре.

attempt number two - what voltage drop on pex reset pin videocard ?

всем привет,

And I have fond an answer. As always it was in front of my eyes.

On the back side, on the left from the notch there are pins 13 and 14 "reference clock" differential pair. Pin 13 is damaged, open circuit. The card is not clocked from the motherboard. I will check traces tomorrow but I think it is a game over.

Thank you very much for help!
Posted after 2 minutes 30 seconds:

After you wrote this I went to measure PCIe bus again.
Thank you very much for pointing me again.
Damn, I checked it several times I measured every capacitor, every lane haha...

well, thanks. it seems I practiced in reballing three gpu's.

Ну, что ж, спасибо. вроде потренировался в реболлинге трех gpu.

This one is broken in my case - it has open circuit reading.

Please measure your boards. Reference clock, lane0 input after caps, and lane0 output, measure 220n caps themselves and of course please measure PEX_RESET.

When a PEX_RESET is buffered then it is a chance that it will not be killed forever.
Anyway if someone has mysterious undetected card please measure it as described above.

они оба измеряют 128 кОм
Мне снова пришлось перебивать мяч из-за того, что один мяч не хотел прилипать. теперь это работает.

they both measure 128 kohm
I had to reball again because of one ball not wanting to stick. now it works.

Please do not measure resistance! Please measure the diode voltage drop. The range is in volts from 0 to 3V sometimes more for some multimeters. The goal is to see if the diode is present or not.

gp104-300 - msi v330 ver 7.1 - 1.567V
gp106-300 - pallit - 1.578V
gp104-300 - Nvidia reference - 1.548V
gp106-400- asus - 1.566V

---

Posted after 1 hour 46 minutes 16 seconds:
very strange is that when using pcie 1x riser those pins are inverse.

очень странно то, что при использовании переходной платы pcie 1x эти контакты инвертированы.

How do you measure 1.5V ? Do you measure between these two pins - I mean between A and B signals? Try to measure to ground. Once black probe of the multimeter to ground and next red probe to ground.

to ground:


gp104-300 - msi v330 ver 7.1 - 0.834V 0.834V
gp106-300 - pallit - 1.578V 0.832 0.832
gp104-300 - Nvidia reference - 0.842 0.842
gp106-400- asus - 0.836 0.836

So now check LANE_0 output both wires and LANE_0 input but after capacitors, also to Ground. And check PEX_RESET to ground.
but remeber that some cards may have an AND gate on the PEX_RESET.

If they all are good it means that not detected may be repaired. It doesn't look like a GPU failure and your chances are higher.
Posted after 21 minute 49 seconds:

I did not analyze it but in normal conditions when the card is good it is not a problem as you see.
The question is what could kill my one pin of the reference clock?

When you insert a riser in opposite direction in a x16 slot then +3.3AUX will go to the REFCLK+. When the motherboard is OFF but the PSU is in standby it can kill the connected card anyway! If you additionally power it ON you will also kill LANE_0 output positive.
We must investigate some more signals anyway...

sorry I don't understand what to measure? the msi 1070 just had some damaged pci interface pins, as in the picture. I was lucky

did you measure you PCI lanes with diode? are they all measuring the same ?

извините, я не понимаю, что измерять? у msi 1070 были повреждены контакты интерфейса pci, как на картинке. Мне повезло

Вы измеряли линии PCI с помощью диода? все ли они измеряют одинаково?

Всем привет читал я тут читал и как бы пришел к мысли а не кинуть ли этот камушек на любую доску 10 серии и глянуть увидет ли компьютор этот камушек. А потом то можно и дальше пообсуждать. Так то жалко списывать в утиль, надо проверять.

Всем привет

Yes, multimeter must be set to a diode voltage drop mode. The diode symbol. You measure from the ground to the pin. For example lane output should give you around 0.4-1V. But not zero and not infinity.
If you want to test inputs you have to measure it on the 220n capacitor on the side of the GPU and it gives mostly 0.2-0.4V.
It must be a finite value as diodes have in the real life. It means that at least some kind of parasitic diodes are present or voltage clamping diodes for protection against reverse polarity, static electricity. If the pin is blown then it reads mostly infinity or around 3V as is the maximum voltage on your multimeter.

Take a picture how do you measure it

---

Posted after 11 minutes 41 second:

This is not only a problem of the 10th series. Every card can fail this way, even PCIe slot on your motherboard. A single pin can be damaged and a dirode measurement will tell you the truth.