Already tried that. No significant improvement over running at uncapped with v-sync (60fps).

baseline:
uncapped with v-sync (60fps)
result: running at 40% usage and 40C temp.

nvidia cp config:
capping at 20fps with or without v-sync (running in background)
result: usage jump up and down predictably between 20% and 50%, no temperature difference.

Clearly worse than capping it at game engine level since ES2 shows linear and stable drop in GPU usage. I assume this varies greatly on different game engine.