Difference between CPU Package power and APU STAPM?

[CatMerc]

I recently bought a Ryzen laptop, and noticed something peculiar in HWINFO. At high power loads, APU STAPM and CPU Package are nearly identical. Meanwhile when the computer is mostly idle, the floor on STAPM is around 2W, while CPU package can drop as low as 200mW.

I tried searching online, and all I can find answering the question is this: https://t.co/bNVzMhsaeM?amp=1

However, its not quite clear. What does APU STAPM include that CPU package doesn't?
If the CPU package reading contains: CPU cores + SoC + iGPU + "others", what is left for STAPM?

More interesting behavior is that when I set my screen brightness to minimum, the delta between battery discharge and STAPM is just 200mW. Surely all the other board components + screen (even if at minimum brightness) draw more than 200mW?

 

[Martin]

Skin Temperature Aware Power Management (STAPM) is a performance enhancement feature that boosts the APU as long as skin temperatures remain below the system-specified ergonomic limits.
Both Package and STAPM power values are estimations of the total APU power level, but they differ depending on the method used to estimate them (i.e. STAPM takes thermal capacitance headroom into account). Those values are then compared against particular limits enforced to determine whether a limit has been hit and performance needs to be reduced.

 

[CatMerc]

Skin Temperature Aware Power Management (STAPM) is a performance enhancement feature that boosts the APU as long as skin temperatures remain below the system-specified ergonomic limits.
Both Package and STAPM power values are estimations of the total APU power level, but they differ depending on the method used to estimate them (i.e. STAPM takes thermal capacitance headroom into account). Those values are then compared against particular limits enforced to determine whether a limit has been hit and performance needs to be reduced.

So is STAPM a sort of synthetic measurement? Akin to how AMD misreported temps to trick its own boost algo on X processors.

 

[Martin]

There's always something synthetic and something real in those measurements. All CPUs/APUs have multiple voltage rails and only some of them are truly measured, the others (not so critical) estimated.
Moreover in such highly dynamic systems it's difficult to make useful measurements as the values fluctuate heavily, hence certain averaging and filtering needs to be applied.
For optimal performance it's often not desirable to control the entire system based on instant values (peaks) and often there need to be other limiting or relaxing criteria applied.
If you drive on a lane with 80 mph limit and need to overtake a slower car, you don't expect the police to shoot you if you speed up to 81 mph for 3 seconds.