Alternatively, I used zenstates and played around with VID values on different p-states (or is it c-states? I can't remember) which allowed my cores to go from ~400Mhz and then speed up as necessary.

There's info scattered around on the internet for how to do this; I don't have my script to do this easily accessible atm.

I do concur on the governor's though, nothing else really makes much of a different. My old HP laptop I permanently had on the powersave scheder and it would draw about 2.5W - the battery life was stupidly ridiculous. My new AMD Lenovo I can't even get below 6W by the same method unfortunately.

I haven’t compared power consumption for a full cycle of active use at 400MHz flat and 1.2GHz flat. I should try it. But as to whether it’s worth reducing the idle frequency, I doubt it: I observe no measurable difference between the two frequencies at idle—though it’s made harder by the fact that the /sys/class/power_supply/BAT0/power_now figure seems to take some time, even minutes, to reach a steady point, e.g. slowly drifting down from 7W to 6.6W. And the lowest it’ll reach seems to vary a lot, e.g. sometimes it won’t drop below 6.6W, and other times it’ll settle to 6.1W, about the lowest I’ve measured. But at 400MHz flat it can’t quite get it to 9W with a dozen or so `while True: pass` busy loops in Python, while at 5W (1.2GHz) I can get it up to almost 11.5W. (And then I throw it into boost mode and watch it climb rapidly to 64W while remaining dead silent, and wonder why the fan isn’t roaring away as it’s supposed to in boost mode and what of my fiddling just now prevented it from doing its job.) Even these comparisons aren’t great because of the lack of GPU load. Wonder if I can throttle that at all?