I really think you aren't reading what I'm writing. I said nothing to you about powermetrics. I've been trying to tell you about alternate explanations for system behavior which occur to me because I actually have professional experience in designing pieces of SoCs (not Apple's, for what it's worth). You seem fixated on thinking all results showing a Max with worse battery life must be explained by the GPU, and invent all kinds of excuses to avoid thinking about other reasons.
Power gating is a real technology which SoC architects use extensively because leakage (constant background power draw in any powered circuit which is independent of clock frequency and switching rate) is real. If all blocks in a giant, complex SoC like M1 Pro or Max were allowed to remain powered on at all times, leakage power would destroy battery life.
That is why I'm guessing Apple power gates a bunch of GPU cores while the GPU is less than fully utilized. But this "guess" is a bit like "guessing" that water is wet; it would be shocking if they aren't applying the most effective power-saving tool in the box.
That is why I keep trying to tell you the large GPU isn't likely to be the root cause of worse battery life on M1 Max in loads where the GPU is not in use. The far more likely cause of across-the-board higher power in M1 Max is the circuitry which has to stay on regardless of activity - such as the two extra LPDDR5 packages and their I/O.
P.S. powermetrics is likely to be highly accurate, but it only reports power sensors for on-die SoC blocks. There's a ton of power sinks in the computer it doesn't measure. The two likely to be relevant to your test are the display backlight and the DRAM. Did you do any testing to figure out how much power the backlight seems to consume? At the setting you were using during your Unigine Valley test? And since it's a zoned backlight which will draw more power for bright images, did you make sure to try both mostly-dark and mostly-bright images to explore that variable?
I bet if you take the time to investigate and subtract display power from system power, you'll discover the leftover discrepancy between that figure and powermetrics-reported "Package power" is small enough to plausibly be miscellaneous components on the motherboard plus the memory.
