Asus F5GL weird behavior

[emestee]

So I pulled an old F5GL off the shelf that I expected to be a donor board and started poking around (a separate "go thank yourself" to Asus for not releasing schematics). For some reason I assumed the board was dead but turns out there was liquid in the onboard power on button. I removed the button and started the board. It boots.

If the OS enables the GPU, the board will hang. Occasionally under linux the noveau driver will show a "GPU locked up" error, and will eventually reboot the board by watchdog; however it will hang in all other cases. I tried several Linux kernels, BSD and Windows 10, the result is the same. If the screen stays in text mode, no problems occur and the board works normally.

Here's the weird part. There are two 1.08V coils on this board. One of them shows 4 ohm to ground, the other 1 ohm. Nothing about this seems normal, however the board seems to work, except the GPU locks up. It also seems the GPU overheats, to the point the board will go into thermal protection if the cooling system is not attached. At this point I am fairly sure this is a dead GPU, but I was wondering if people with far more experience encountered something like this and if it has better explanation.

[piernov]

Re: Asus F5GL weird behavior

Yes that sounds like a dead GPU.
Post a picture of the GPU die.

[SMDFlea]

Re: Asus F5GL weird behavior

Boardview + repair guide, if it helps - https://www.badcaps.net/forum/showthread.php?p=1007062

[emestee]

Re: Asus F5GL weird behavior

Thank you esteemed elders!

This is the GPU: https://photos.app.goo.gl/T4ndjYckVmqzHn7P9

As you see the factory fixing compound is intact. The protective film appears to be bubbled up slightly. The bright areas around the edges are reflection, the white matter around the crystal appears to be thermal compound that got under the film. Unfortunately the board is too big to examined under microscope.

Edit:I'm retarded, I forgot to flip the board in boardview... Bottom line: +1.05VO is 0.4 ohm (was 1 yesterday, what now!!); MCP_VDDO is 4 ohm.

[piernov]

Re: Asus F5GL weird behavior

Ok this is an NVidia MCP so northbridge/southbridge/iGPU in one chip. This one should not be affected by bumpgate (and has the good "white" underfill), but they can still sometimes become unreliable after several years.

Note that when measuring resistance to ground, board has to be dead cold and all power drained from it. If +1.05VO was really shorted to ground the machine wouldn't turn on.
I don't know what a typical resistance to ground value is on this chip though.

[emestee]

Re: Asus F5GL weird behavior

The board _was_ completely turned off when I measured. I can pull out the RTC battery out of curiousity, but I dont think this will change anything. However the fact remains that, at half an ohm, the 1.05 rail is being fed into the MCP through at least 3 different pathways (hence the near 2A consumption even when the GPU is not being used). I suspect that one of these is GPU core, which, while shorted, is not being enabled until an OS driver enables it, at which point it hangs. It's entirely possible that the MCP has another video core and they switch; I will boot Linux in textmode and see what shows up on the PCI bus probes.

I'm going to pull out the jumpers one by one to see which one would cause the resistance to go up. I will retest everything again but it's still true that MCP_VDDO is 4 ohm and 1.05VO is 0.5 ohm.

For the record the board has absolutely no value to me and I am only doing this to satisfy my curiousity.

[emestee]

Re: Asus F5GL weird behavior

After several evenings of mucking around, having to open up my PSU to adjust the voltage indicator pots so that screen isn't 1V off, and a lot of moving of caps and coils, and removing jumpers, I finally found the culprit. Because this was in a 1.1V line, I had to be very patient. It was a tiny decoupling cap tucked away almost under the PCH frame.

Now the 1.05 rail is 200 ohms instead of 0.4, and the MCP_VDDO is in tens of kiloohms. The board powers up and seems to be able to make a better progress, although ultimately it still hangs. The current consumed by the injection was instantly halved. When the machine is on, the MCP does not heat up as terribly as it used to. I will continue probing it.

The lesson I learned from this exercise is: even at low voltage lines with sensitive stuff like the PCH nearby, if you raise the voltage very carefully above the nominal and see a current spike from the PSU, chances are your can find the short with your finger. In my case, the cursed cap dissipated maybe 2W of power, but it was enough to locate its treasonous ass.

Of course, I tipped the board a little after removing the cap and it escaped. I don't count on ever being able to find it and execute it properly. But the learning experience was totally worth it.

By the way, this board turned out to be designed for a separate 1.1V power supply, but the lazy cost cutting manufacturer just removed it and tied 1.05V line into it directly.