Trying to fix ZOTAC RTX 3080

[pastapoulos]

Hi guys,

I bought a broken GPU that I'm trying to fix. The card powers on but isn't detected by the motherboard.

The card is a ZOTAC RTX 3080 Trinity OC 10GB. I can't find any schematics for this model, so it's a bit challenging for me to diagnose it with my limited knowledge.

I haven't found any shorts, and the resistance measurements seem normal.

Upon visual inspection, I noticed an inductor with a small crack (see photos), but I'm not sure if it could be the cause of the issue. Do you think I should replace it? Does anyone have more information on this component?

Thanks in advance, guys!

[R_J]

That small chip on that inductor is not an issue

[pastapoulos]

Thank you very much, I will continue my investigations.

[pastapoulos]

Finally, I'm facing an issue with the 3.3V line. When measuring the 3.3V pin in diode mode, it indicates an open line. I also measured pins 1 and 3 of the PEX_RST component, and I'm only getting 1V instead of the expected 3.3V. How can I identify which component is causing this issue without access to the schematics?

[pastapoulos]

Finally, the 3.3V is indeed present. It was probably a measurement error. I took several measurements again. There is an issue with the vcore power supply, but I don't know where to look.

[rogfanther]

Measure the coils of the Vcore supply. Any short ?

If not, and only 0V, check the IC responsible for this rail. Which is it, and does it have the correct voltage on its VCC pin ?

[pastapoulos]

When you say measure the coils of the vcore supply, you say resistance to ground ? Resistance is very low about 1 ohm.
I have 5v on VCC of the controller

[rogfanther]

Ops, sorry. Yes , meant resistance to ground.

[piernov]

No point in measuring resistance to ground of GPU VCore with modern GPUs, it will mesure close to 0 ohm.

[pastapoulos]

Hello, the only thing I notice is the absence of NVDD_EN on the U812 circuit. According to a YouTube video by Northwestrepair, this signal is generated by the circuit itself. When I measure the signal on pin 36 and power up the board, it ends up being short-circuited. In this video, they recommend desoldering a dual NPN to force the signal and then powering the GPU with a DIY power supply to look for a MOSFET issue using a thermal camera.

Do you think this is a viable solution? I have a lab power supply; could I potentially diagnose the issue by injecting voltage? What would you advise?

[DynaxSC]

Do not fully agree that measuring GPU VCore to GND is no point. The issue is however that these resistances are really low, 1/10 - 3/10 ohms, but with a suitable low ohm meter it is possible to measure this very precisely. There are special low ohm meters that work in a bridge constellation with 4 wires from meter to probes. This type of meter corrects the wrong readings resulting from probes wires resistance, and can measure the low resistance very precisely. I use such a meter with very good results on GPU's, telling me if the core is generally shorted or not. This is a very good and fast test of the GPU Core, when I receive a fresh broken card telling me if there is a chance to have a healthy core. There are such meters on Aliexpress available, and they are not very expensive. A good GPU Core resistance is not a warranty for a fully healthy core, but usually it happens it is healthy. Even more, using such a meter it is possible to determine if a short is on side of the GPU or on side of the power mosfets, without desoldering the coils. So this saves a lot of time for diagnostics. With luck it is even possible to determine which phase is shorted.

Regarding the injection of power to determine shorted mosfets/GPU you can try it, but the difficulty is that a shorted mosfet or GPU have usually almost no resistance, meaning very weak or even no thermal effect. In such a case you need a very high current output, usually 20-30A or more to have some effect you can see on the thermal camera. However it depends on the specific case and the specific rest resistance of the broken part. There is also a risk, that, assumed the short is on the mosfets, if you manage to accidentally burn away the short, the voltage will rise above GPU specs and will destroy the GPU. So you must be sure the voltage is limited below GPU specs, or much better desolder the GPU or all coils before doing this.