GTX 1070 Ti - TDP Issues - Always Power Throttling

[Th3Knights]

I have an ASUS Strix GTX 1070 Ti. The card always shows a TDP of over 200% which causes the GPU clock to throttle down to 139MHz and Memory Clock to 405MHz. I have tried using the card in 3 separate PC's, tried various NVidia Firmwares, and have tried to reflash the card using ASUS's latest vBios for the card. (It has not been vBios flashed before but I wanted to rule out a bad firmware)
I have come to the conclusion that the fault is with the card itself, specifically hardware related. I have disassembled the card and thoroughly looked for any potentially burnt or bad chips and could not find any.
I'm including an image of FurMark running a benchmark only getting 5 FPS at average. I'm also including a GPU-Z and HWInfo64 screenshot during those runs.
On the bottom left of FurMark, it says the TDP is 506% with very low Core and Memory speed. While GPU-Z shows the TDP as being 207%. Then if we look at HWInfo, all the power lines are wrong. I suspect the issue is with a chip but I'm not sure which chip would control power readings. I suspected the VRMs, as the 1070 ti has 4 VRMs on the board but HWInfo only had a reading for 2. Not sure of what to think. Any ideas would be appreciated.

 

[Calenhad]

It is trying to pull over 300W from the PCIe slot, and close to nothing from the 8-pin input. Do you have a proper 12V input to the 8-pin connector? If yes, I'd say something is very wrong with the 8-pin power delivery for starters. And trying to pull >300W from the PCIe slot if there is no 8-pin 12V connected would also be seriously wrong.

So either way that gpu have issues. Big issues. My immediate suggestion would be to recycle that one and find something newer and better.

 

[Th3Knights]

I've tried it with 3 different PSU's and they all have the same results. Its seems salvageable since it at least powers on and I'm able to use it. It doesn't have any distortion either. I didn't realize it was trying to pull the 300W from the PCIe port, I thought it was the total power the board was receiving but I see I overlooked it. If you had to guess, would you say the chip that is causing the issue is near the 8 Pin connector or the PCIe connector?

I do have a separate build that I use with a 7900XTX, this was just something I had laying around that I was going throw in another PC that I have. I won't be too disappointed if I have to replace it, but I'm OK with working on logic boards so If I were able to find the bad chip/mosfet/capacitor, I could replace it.

 

[Vya Domus]

The circuit that's measuring power on the board is busted, it's a hardware fault, doing a shunt mod it's probably your only option.

 

[DR4G00N]

The circuit that's measuring power on the board is busted, it's a hardware fault

This is correct, I believe it's the INA3221 monitoring chip that's failed. Replacement chips are readily available and fairly cheap so you could probably have a go at replacing it and seeing if it fixes it.

 

[Th3Knights]

I was able to replace the INA3221 chip from an old GTX 660, the 1070 Ti still powers on but I'll have to hold on testing it. I've ran into another issue with this secondary PC not wanting to post. (Irrelevant to the 1070 Ti as it won't post even without it.) Once I get that working, I'll report back on if that chip change made a difference. Thanks for the suggestion!

 

[Th3Knights]

I wanted to provide an update. Changing the INA3221 chip didn't fix this issue, however, it did get me looking more into that area of the board. There are 2 20 ohm resistors near the INA3221 chip that receive a signal from a shunt resistor near the PCIe connector. I tested them both out and one is 20 ohms and the other is 114 ohms. The schematics of the board also show them as both supposed to being 20 ohms. I'm going to be replacing them both with 2 new 20 ohm resistors and see if that resolves the issue. Evidently, the shunt resistor transmits information about the PCIe power consumption that passes through those 2 20 ohm resistor which is sending a bad reading to the INA3221 chip causing the GPU core to cap at 139 MHZ. Unfortunately the GTX 660 doesn't have any 20 ohm resistors so I had to order them.

 

[Th3Knights]

Another update:
I changed the 2 circled 0402 resistors with 20 Ohm ones and tested the card out and ran a benchmark and the card functioned normally. However, after 2 or 3 minutes of stress testing, the results fell to how they were and the original issue came back. I went back and checked both of the resistors and the middle one had changed from 20Ohms to 40Ohms. This resistor in specific goes to the 12V_INP on the Shunt Resistor by the PCIE connectors. I removed the Shunt Resistor and wasn't exactly sure how to test it as it measured the same as other shunt resistors from other boards. I put it back on the board and swapped the 20 Ohms resistor that went bad with another one. Booted the PC and adjusted the card power to 90% while slightly lowering the core frequency's max. Ran another stress test for several minutes and it seems to be working fine now. I'll probably fiddle around with the power settings and see if I can go higher without it failing again. If anyone would want to try to take a guess at what might have caused that resistor to fail after replacing, let me know. Thanks!