Gainward RTX 3060 12GB with unusually low VCore

[Nick Nougat]

Hi, I'm currently trying to diagnose a Gainward RTX 3060 12GB "Pegasus" that I bought as damaged. I know nothing of its prior history.

A boardview for an MSI model, which matches the board layout apart from some resistor values I saw, was already uploaded in the past if you'd like to have a look (I used file V397_70). See www.badcaps.net/filedata/fetch?id=2101580

It does not have any shorts but it's not detected by the system (i.e. not even shows up in the output of lspci and the CS pin of the Flash chip is never accessed, so apparently the GPU does not boot..

After checking for Voltages, most seem normal, i.e. ~900mV PEX, 1.3V VRAM (FBVDDQ), 5V, 1.8V. However I see only a very low 304mV on the VCore (NVVDD) VRM coils.NVQDD_PGOOD on the PWM chip (uP9512R (Data Sheet: https://www.upi-semi.com/files/2461/...f-d30ecdd7d08d), U518, bottom side) measures high (and explains the presence of the PEX and RAM voltages).
I checked for phases on the VRM coils with my cheap Hantek 6022BE oscilloscope and I see spikes on all of them, so they all seem to be active.
I also checked the voltage input of the NCP302045 integrated MOSFET+Driver combo chips on the rail and they all get 12V.

After digging a bit into how the uP9512R chip works, it seems that it should get a PWM signal on the VID pin that allows the GPU to control the desired voltage, which in turn, via REFADJ will set the desired VRM voltage at REFIN.
Measuring REFIN gives me 304mV, so the VRM seems to be working but appears to get a wrong desired voltage.
Probing VID gives me a constant voltage of around 222mV, not a PWM signal. According to the schematic, VID is routed directly to the GPU, which makes me wonder if the GPU is damaged or loose.
To rule this out I also tried to check whether the crystal provides a stable frequency, but my measurements were inconclusive.
Given that my cheap Hantek 6022BE can only measure up to 48 MHz and the crystal has 27MHz, this might be due to sample rate issues.
What I got measuring XTAL_IN and XTAL_OUT to GND was a ripple at ~900mV that basically looks like this:

I would have expected some kind of much more prominent waveform. The curve is very similar to the pretty high noise floor of the Hantek.
Since I have no real experience measuring oscillators/resonators, I would appreciate feedback if I did it right. I measured C213 and R213 to GND.

So what do you think, is the GPU damaged or the crystal, or am I overlooking something fundamental?


Thanks very much in advance!

[Nick Nougat]

I just realized that embedded links don't seem to work, so here goes another try with URLs.

Link to schematics: www.badcaps.net/filedata/fetch?id=2101580
uP9512R data sheet: https://www.upi-semi.com/files/2461/...f-d30ecdd7d08d

[Nick Nougat]

Small update: After looking closer at the VCore voltage, I noticed that at the very start at power on, the VCore is at around 700mV, which I assume is the standard boot voltage, but drops to 300mV after a few milliseconds. The REFADJ pin also has a similar voltage around that time and then also drops to 0V.
Other than that, since the BIOS chip had pretty obviously been removed and soldered in again rather crudely, I decided to check the BIOS and found out that it definitely wasn't the original, so I replaced it with the stock ROM to make sure this wasn't the root cause after all. Unfortunately this didn't change anything.

I'm running out of options, so any help is greatly appreciated.

[Nick Nougat]

OK, I did some more looking around and I managed to analyze a similar crystal on an old MSI R9 380 I have which has defective RAM, but a working core.
It's surprisingly hard to get good info on how 4 pin crystals work and what voltage and signals to expect on the pins.

On the R9 380 I have around 850mV to GND on both pin 1 and 3 of the crystal, which I assume is a credible RMS for a sine at 1.8V.
According to a schematic for another R9 380 model I am assuming that the crystal has an identical pinout to the RTX 3060.
When I touch a probe to pin 1 I can see a sine wave on my oscilloscope at about the expected frequency, but only if I don't ground the probe. If I do, the signal is flat. The crystal GND does not seem to be connected to GND, though, so that might explain that.

On the RTX 3060 I only have around 860mV to GND on pin 3 of the crystal. Pin 1 is at 0V. I don't see a sine signal on any of the pins, whether I ground the probe to GND or the GND pads of the crystal or not.
On the RTX, the casing and the GND pins are connected to GND while on the R9 neither of them are.

My current assumption is that the crystal or the circuit driving it is defective. This would explain why the GPU is not taking over after getting an adequate boot voltage and why it does not generate a PWM signal for the buck controller.
However, since I have no real knowledge (yet) of how crystal oscillators in GPU circuits work I am not too sure how to proceed.

I looked at several crystal datasheets but none show the any typical application circuit that could help me get some basics down.
Can someone point me to resources where I can learn about how crystals and how they work in a typical GPU circuit setup?

Alternatively could someone answer these questions:
- In terms of voltage and frequency, what can I expect from pin 1 of a crystal that is connected to XTAL_IN on the GPU?
- In terms of voltage and frequency, what can I expect from pin 3 of a crystal that is connected to XTAL_OUT on the GPU?
- What typically is XTAL_OUT_BUFF on an NVidia GPU? Is my assumption correct that this is an amplified xtal signal? I can not find any application of this in the schematics I have. It only appears in one place in the schematic along with mentions of voltage levels at different PWMs.

Thanks for any pointers!

[Nick Nougat]

Since nobody is reacting to my post, I'd appreciate to know what I'm doing wrong here. The last thing I want is to annoy anyone or waste your time (or mine for that matter 😉 ). So if anyone could enlighten me, I'd really appreciate it.

In the meantime I have been trying to find more info on the crystal situation and after a long search I finally found some basic info on oscillator/resonator setups that leads me to believe that I am dealing with a crystal here.
I am ruling out that this is an oscillator because, among other things, it has no own power supply.

With this information I'd answer my own questions from the previous post as follows: The GND pins on the crystal are for shielding only and comparing the voltage levels to my R9 with a working GPU chip, I'd expect the RTX crystal to have roughly the same voltage levels on both XTAL pins.
Pin 1 of the xtal measures open to GND, which I wouldn't expect for a GPU input pin (and on my R9 it reads a few MOhms), but which explains the 0V reading.
Since my R9 also has a 27MHz crystal, I decided to make an experiment and swap the crystals to rule out that this is the root cause. Unfortunately this did change nothing in the behavior or readings for the RTX.

All this seems to support my initial assumption that the GPU is probably dead and has no more capability to drive the crystal - I'd really like to hear your thoughts on this and/or any other ideas I could still try if I'm wrong.