P150SM - EC Panic related to GPU

Hi,
a few months ago, without any soft/firmware or driver update I can remember of, my good old P150SM started to go on EC panic (fans to max, keyboard LED flashing, loud beeping, followed by shut down) when I was playing. Sometimes even running GPU-Z to monitor the GTX 860M is enough to start the EC panic. The time I can use my GPU is random. Can be 30 seconds (most of the time), can be 5 hours (rarely).
Quiting the game or GPU-Z switches back to the integrated GPU (i7-4700MQ), and the EC panic stops.
I guess the EC is asking for some low level information, that the GPU fails to deliver (which could be temperature, considering the fan activation). Windows-accessible GPU temperature is fine (below 70°C), but at the EC level it might not be.

Not being able to play is a problem, but there's a more serious one: before windows switches to the intel integrated GPU, the EC wants to know how the NVIDIA GPU is doing, and that means EC panic while I'm accessing BIOS, booting on a memtest stick, or trying to install another OS (currently running Windows 7). That really sucks, and it could even shut down before finishing loading windows, would the issue worsen.

I tried to unplug and replug the GPU, in case contacts were bad. I didn't dare to "clean" the contacts, as they look pristine and didn't believe my isopropanol-drenched cloth would make any good. I haven't baked the card either, it's kinda scary.

I tried to run my PC without the GPU (integrated GPU takes charge), but it's guaranteed EC panic, although I have no display problem whatsoever, and even after windows is running.

EC firmware is up-to-date with clevo stock, I think (1.03.01)

What solutions do you guys see?

 

Below 70c in what case, at idle? Does the fan spin up? What other behaviour is the system exhibiting?

 

Yes at idle with fans still quiet. If I put the fans to max (fn+1), it will decrease under 50°C.
I don't see any other unsual behaviour in the system... As long as I don't execute a game or stay in BIOS for too long, everything works fine.

 

70c idle does suggest an issue.

 

Either your thermal paste is really bad, or your heatsink has a leak, because 70ºC idle is way to hot.

 

Hmm, maybe these 70°C were just after I put back the GPU, it was my first repaste and I later saw that the die wasn't completely covered by the paste. I've redone it since then and now idle temperature is under 50°C.

The problem persists. Here's a GPU-Z screenshot. It's pretty flat, but near the end of the graph, my PC started beeping and flashing LEDs, and if I didn't stop monitoring the GPU, would have shutdown. I've never seen any clue in these graphs that made the EC panic foreseeable. Am I missing something?

 

Unplug the AC adapter for 30s while the system is off.

If that doesn't help reflash the EC.

 

Also make sure your vrms are making contact with the thermal pads properly.

 

It seems the problem was actually from the CPU. Repasting it has almost solved the problem. Redoing it better or with a better paste might solve it completely.
I mostly get the issue on windows 10, because I can't get to install my GPU's driver (it's not even recognized as an NVIDIA device), then the integrated GPU takes all the load and overheat after some time. At least it's my guess.

 

Did you check both VRMs sets are making contact?

 

Good question. I remember specifically checking for the GPU, but I'm unsure for the CPU. The die is a rectangle, I found it somehow harder to repaste than the square GPU.

 

Rectangle = line, square = x paste pattern.

Also check the pads for a vrm imprint.