So... everyone keeps bashing on windows, so I thought I'd restore the balance a bit with my story of how windows is fixing some NVIDIA's driver fuckups which it probably shouldn't be able to fix, and if yes, surely not in this way.
(Copypasted from my post in lenovo support forums, because I'm too lazy to write it again, why would I?)
I have an IdeaPad y700 with the 4gb VRAM version of the NVIDIA GTX 960M dedicated graphics card.
I have been having weird complications with system switching between the onboard intel gpu and the dedicated one probably since day one. Windows usually had them switched around (in Device Manager) meaning when I disabled the one marked as NVIDIA, the laptop actually turned off Intel and ran fully on NVIDIA, and the other way around, the auto-switching didn't work properly in games, or even the manual switching set up in NVIDIA Control Panel (hence my experiments with explicitly disabling in Device Manager), etc etc.
After I solved that, everything was fine, until some most recent issues with throttling due to overheating, which is fully my problem, yes, I'm working on it, going to disassemble and clean and replace the heat conducting paste today, also finally got a cooling pad...
BUT... the strange thing is... when I turn on the computer, it immediately starts heating up. NVIDIA Inspector reports dedicated gpu usage at 100% and temperature starts at 65C and is rising immediately. Which makes sense for 100% usage with very clogged fans... However, that's right after startup, nothing going on, just my windows desktop...
and then, the weird thing is (for some time I thought it's only placebo, but it's not) that when I launch task manager, as soon as I switch to the "Performance" tab, in its usage graph, both NVIDIA and Intel GPUs spike to 100% for a fraction of a second (I assume as there's memory copy happening between the two), and then NVIDIA shuts off (in task manager as well as NVIDIA Inspector), its usage goes to 0, and task manager goes on reporting Intel gpu's usage on 2-6% (normal values for desktop usage).
Sometimes the switch causes the usual screen going black for a bit.
Then, after a while, NVIDIA Inspector temperature graph drops to zero as (I assume) the power to dedicated gpu is completely shut off, including the temp sensor. As it should be and is expected.
I found this Task Manager effect by complete accident, originally as a solution to FPS drops in games due to throttling (or just driver being confused, not sure which was first), as keeping the Task Manager open on the Performance tab solves these drops (even now when temp throttling is happening), resulting in silky-smooth framerate.
I have found out the rest of this behavior (with pc properly switching to intel gpu when nothing intensive is running) later.
I am not kidding you, something that specifically Task Manager's Performance tab does on its initialization causes the GPU switch the way it properly should, but doesn't by itself. But the Performance tab does some magic/flips a switch that makes it work.
Any... explanation, at least? Or hypotheses? Or... anything? Would be welcome, because... I'm glad I found a way to "solve" the long-running complications with the GPUs, but... I'm bewildered and I need at least some kind of answer... =D
Also, yes, over the time, months of observing and experimenting with this problem, I used NVIDIA Inspector to mess around with driver flags/settings (but not overclocking/underclocking section) in various ways, but it seems the effect of Task Manager's Performance tab persist across any settings I choose in there.
I'm running Windows 10 that just updated today, and latest NVIDIA drivers, but this issue has been going on across many, many updates of both OS and GPU drivers...
(Edit: tag reordering ruined my point there :( )