Starfield indeed stands out in my experience. And others seem to have had similar observations. It’s possible that it uniquely hits particularly unoptimized paths in the driver. Hopefully, debugging these will lead to broader improvements in driver efficiency. Yet, with the complexity of modern game engines, I can also easily imagine different factors—and in different combinations—being at play for each game.
As for the impact of resolution and graphics settings, that does seem to be an established pattern—also supported by my results (e.g., ~84% relative Linux/Windows performance on Ultra in AC Shadows vs. ~68% on Low). This pattern might suggest that the driver isn’t fully utilizing available GPU resources when the workload is lighter—perhaps due to overhead becoming more noticeable relative to the actual rendering work. It’s not my field of expertise, so I can’t speak to the specifics, but the general concept seems plausible in my view.