How I Test
Every number on this site came off my own laptop. Nothing here is pulled from a spec sheet, a press release, or somebody else’s review, and where a claim comes from documentation rather than my own testing, the article says so in that sentence rather than leaving you to guess.
The test machine
Everything gets tested on one machine, an ASUS ROG Zephyrus G16 from 2024 running a Ryzen AI 9 HX 370. I bought it refurbished, which matters more than it sounds, because it arrived at 90% battery health and that is what got me into measuring any of this in the first place.
It runs Windows 11 25H2, and articles name the exact build when the build could plausibly change the result. It is also the laptop I actually use for class and coding, not a test bench that gets wiped between articles, so what you are reading is how these settings behave over months rather than over an afternoon.
How things get measured
The tools are deliberately boring and free, so you can repeat any of it yourself:
- Battery health — the built-in Windows report, run with
powercfg /batteryreport - Power draw — G-Helper’s own discharge readout, watched over real working sessions rather than a single snapshot
- CPU performance and stability — Cinebench 2024 multi-core, plugged in, run at each setting being compared
- Memory usage — Task Manager, measured with a feature on and then again with it off
- Thermals — the temperature readouts in G-Helper under the load being described
Where the limits are
This is one person with one laptop, and it would be dishonest to pretend otherwise.
Silicon varies, so an undervolt that is stable on my chip may not be stable on yours, which is why the undervolt guide walks you through finding your own number instead of handing you mine to copy. Results on a different model, a different Windows build, or a different BIOS version can land somewhere else entirely.
Where I have not personally hit something, the article says that outright. A line like “this comes straight out of the official guide rather than something I ran into myself” is there on purpose, because mixing research in with first-hand testing and presenting it all as experience is exactly what makes most guides useless.
When something turns out to be wrong
Articles get updated when Windows moves underneath them, when a tool changes, or when a reader points out something I got wrong. Testing that contradicts what I expected stays in the article rather than getting quietly dropped, because a theory that did not survive contact with the hardware is usually more useful than one that did.
If you have run the same test and got a different number, I would genuinely like to hear it. That is the fastest way this site gets more accurate.