I have an MSI GE62-2QD laptop running Windows 10 v1607 (build 14393.0) with an Optimus graphics configuration (GTX960m discrete and Intel HD Graphics 5600 on an i7-5700hq).
Occasionally, the laptop encounters overheating (about 90 degrees) and begins to throttle by limiting CPU frequency to about 800MHz, until cores drop to 89 degrees. However, when throttling occurs (for example in a game that is using the NVidia GPU), the Intel XTU utility reports that the power consumed by the Intel graphics coprocessor jumps dramatically, from 1-4W to about 18W. This additional power consumption causes the temperature of the die to continue to rise, making it hard to break out of the overheat situation.
I've already tried to clean and replace fans, as well as re-seat the heatsink with new MX4 thermal paste (cleaning all surfaces with the recommended cleaning processes), which has kept temperatures slightly lower but still causes occasional thermal throttle events.
This additional Intel GPU power consumption does not make sense, since I'm not running any GPU-intensive processes that utilize it, and it shouldn't be receiving any additional load. According to Intel XTU, the graphics frequency is as low as 200MHz during the throttle events, and running GPU stress tests on the Intel GPU doesn't lead to such high power consumption when the system is not throttling. Additionally, on Linux (dual-boot), the system recovers from throttle events almost immediately, leading me to conclude that the issue is not present at that time.
I've attempted to use Throttlestop's BD PROCHOT flag (as well as other parameters such as max graphics frequency, graphics undervoltage, etc) to no avail. The NVidia GPU (unaffected by this) has no cooling problems.
Is there any known fix for this?