GTX1060 not working, Vcore down to 0.3V

[stephtech]

Hi,
Need some help as I am struggling with this GPU.
It is an Asus Dual GTX 1060 3GB

Current problem when I received the board (I do not know previous history), device manager show error code 43. GPU-Z show UNKNOW in Bus Interface (PCIe speed), no display. GPU-Z show all the infos for the board, only unknown for the bus interface.

Steps I took:

1- Disassembled the gpu and physically inspected everything on the board. No missing components, no shorts, no burnt marks, all looks normal. No open 0 ohms resistors.

2- connected the gpu in a powered riser but with the USB cable that goes to the motherboard not connected, basically, just power. I then turned on the power supply. Validated all the voltage rails are fine, Vcore=0.81V, Vmem=1.5V, 12V, 5V, 3.3V, 1.8V & 1.0V, all seems ok.

3- Repowered on the GPU but this time with the riser plugged in the motherboard. When the computer initialize the video card, all voltage stays the same but the Vmem goes from 1.5V to 1.35V and the Vcore goes from 0.81V to 0.31V. Computer beeps error as it cannot find the video card. If I press the reset button on the pc, Vmem and Vcore goes back to 1.5V & 0.81V for a few seconds, until the video card gets initialized again.

4- I then thought it might have the wrong bios, tried with nvflash but it said no Nvidia card found. I then downloaded the bios for this gpu and flashed it to the bios chip successfully with an external programmer. Flash chip seems like working normally, read/write/compare shows that it is stable. then retried with the pc; same behavior with the Vmem and vcore dropping.


I might be wrong, but I think that once the video card is initialized, the Vcore should drop to somewhere in the 0.7V, if so, my 0.3V is way too low and must be related to my problem.
I have the boardview for this board and a close enough schematic.

any clues/tips would be really appreciated.

[Bozog]

Have you tried running mats memory test?

[stephtech]

Originally Posted by Bozog

Have you tried running mats memory test?

yes I did but no GPU found

[foshland]

Does the voltage drops without the riser? direct in PCIe slot?

[stephtech]

Originally Posted by foshland

Does the voltage drops without the riser? direct in PCIe slot?

yes, same behavior

[dicky96]

Whatis the part number of the Vcore controller IC?

[stephtech]

Originally Posted by dicky96

Whatis the part number of the Vcore controller IC?

It is a UP9511Q, not the usual P which has more phases.

For the up9511Q, there is not datasheet available but I did found an image with the pinout.

You suspect something with the controller?

[DynaxSC]

Just an idea. Check if the vrm has a serial resistance in the 5v supply line, usually they are 2,2 ohms, sometimes they loose parameters due to overheating (bad design, to weak wattage parameter), and can have a higher resistance, so the vrm does not get the full 5v.

[DrasticRepairs]

It's probably the crystal oscillator. Check for 0.9V and a sine wave from both sides of the crystal.

[stephtech]

very good quality suggestions here;

For the VCC on the up9511Q, there is a 2.2R resistor inline and it is ok, 4.97V steady at power up and after bios try to initialize the card.

As for the crystal, I do have 0.9V on XTALOUT and XTALOUT lines. Will dig a scope tomorrow to confirm proper waveform.

[stephtech]

* did checked the crystal and waveform are nice and steady at 27Mhz *

While there with the scope, I probed the bios chip on pins 2 and 5 to see activity...not much going there but I was using a very old scope that refreshed way too slow to be trustable.

will validate with my Logic8 to get real infos.

[victorioz]

I had same problem with my Zotac GTX 950. I've been change the Voltage regulator (up1503V), 5v regulator, Crystal oscilator, and GS 7103. But the problem still appear.. Hope you'll find the problem.

[DrasticRepairs]

If the xtal is good, it would have to be something with refin on the up9511. So check voltage on refin when the core drops. If that's the same as vcore, you can eliminate any power source as the problem. Then check the pwm on vid, and the voltage on vref, and refadj. Also check all resistors between refadj, Vref, and refin.

[stephtech]

Originally Posted by DrasticRepairs

If the xtal is good, it would have to be something with refin on the up9511. So check voltage on refin when the core drops. If that's the same as vcore, you can eliminate any power source as the problem. Then check the pwm on vid, and the voltage on vref, and refadj. Also check all resistors between refadj, Vref, and refin.

I think you are on something!
the Vref pin is showing 1.85V and when the Vcore drops, it is around 1.0v and not stable. According to the datasheet for the P version (should be the same for the Q), it is supposed to be a stable 1% 2.0V reference.

will have to dig deeper tonight what is going on, to see if the 9511 is acting up or if it is an external component. If internal Vref is screwed up, I think I will just temporarily fit an SOT23 2.0V regulator to confirm proper operation.

will update later with my findings.

thanks for your time and suggestions!

[dicky96]

Yes that is what I suspected - some problem with Vref, the COMP voltage or the Feedback resistors that monitor Vcore, but DrasticRepairs got there first

[stephtech]

oh crap, I checked the wrong Vref pin..I checked by mistake pin 8, this is for the 9511P, I have the 9511Q which is pin 4 for the Vref.

So Vref is rock solid at 2.0V, the Refin is actually following the Vcore (from 0.8V to 0.3V)
I reversed the resistor network (because i do not have the correct schematic) and all is good in comparison to the same network of resistor for the 9511P.

what I did noticed though is the VID pin, upon power up, it is at 0.92V solid and when gpu is initialized, it has pwm signal with about 0.9V amplitude and roughly 5% duty cycle. I am not sure if the shape looks right, it has a very long rise time and if I monitor the signal for a few minutes, the amplitude seems to drop a lot, I really suspect my crap 20years old scope... Right now, with that crap scope, I cannot get a clean signal. I will grab my other scope and get a real signal, then I will know if this signal is distorted or not

I originally thought that there was no pwm signal present but realized I had forgotten to plug the PCIe power connector...

Resistors connected on the 9511q pin 1 (vid) that are on that line are 0ohms resistors and they are fine. The rest of the trace then disappear and I cannot follow it.

I guess the next update will be when I get my good scope, unless you think of something obvious based on these infos.

btw, thanks guys for the time you take to help me...really appreciated!

[stephtech]

Originally Posted by dicky96

Yes that is what I suspected - some problem with Vref, the COMP voltage or the Feedback resistors that monitor Vcore, but DrasticRepairs got there first

I just realized that I missed the COMP to verify also, I just did.
Upon power up, I got 1.59V and when initialized, I got 1.43V

not much information in the datasheet for this signal as to what expect.
Does those voltages seems ok?

[stephtech]

So I may have found something (with a nicer scope), the VID signal going to the 9511 is problematic.
It is not always present, when it is, it is around 0.9V in amplitude and frequency is varying a lot. Rise time is also not sharp, it is a really crappy signal at the moment.

from the 9511 datasheet, it specs 1.2V minimum to detect a high signal on this pin, I got less than that. The Refadj which mirror the VID pin, goes from round 1v to 0 when initialized, thus proving that it does not detect any high level on VID pin.

Now, question is: is it the output pin from the gpu that is problematic or something in between or the 9511 chip itself. I only have the 6Gb ram version of the schematic (mine is the 3Gb) and the Vcore controller is not the same, so I cannot tell if the VID pin goes straight to the gpu.

[dicky96]

VID is generally serial data coming from the GPU. This data is loaded into an internal D2A converter in the PWM controller, and the resulting analog voltage output from the D2A generates COMP which is then added/subtracted to/from the FB (feedback) voltage, usually with an external resistor voltage divider... and this sets/regulates the output voltage Vcore

At least that is the way I understand it, though I didn't examine this particular datasheet. If COMP is changing from 1.59V on startup, to 1.43V, it suggests that the GPU has control over the VRM, so VID->COMP is actually working.

[stephtech]

well, I do not want to argue with you but from the 9511 datasheet, it is mentioned that it is a PWM signal, not serial. And also the fact that COMP voltage changes is because the VID signal from the GPU is tristate until initialized, then it modulate the PWM signal. When tristate, this signal is high, when initialized, pwm kicks in and contribute a voltage in the feedback loop. But because the vid signal is only seen as a low, it just lower the output voltage by a factor predetermined by the resistors network. I spent a lot of time reading and understanding this aspect of the circuit.

So your comprehension of the circuit seems about right, only the VID signal is pwm and not serial.

Now one test I can make would be to remove the 0ohm resistor inline with the vid line going from the gpu to the 9511 and validate that the voltage no longer drop but I am afraid that the Vcore might jump too high upon power up and also, I am sure that if this portion is not working, it must mean something else is also in trouble on the board.