The old metric was power draw in Metro Last Light at one resolution vs average gaming performance for all my game tests in three resolutions
This is problematic where cards are CPU limited, especially at lower res. and 6900 XT is much more CPU limited than 3090 for example, which skews the results even more. You can find examples of this in older reviews.
Ideally the test would have been average power draw in every single game per resolution, but that's difficult to record, and analyze.
So now I'm measuring power draw and FPS in Cyberpunk 1440p, which isn't CPU limited on any card, every vendor has optimized drivers, and it reduces the complexity from 3 charts to 1, without losing much relevant information