Interesting that the system switches between two distinct devices at different power/performance points. There is probably some redundancy here, in that the higher power/performance (discrete) GPU likely includes all the elements of the simplified (integrated Intel) GPU. So the discrete GPU could be designed in a way that dials down the power and performance to match the discrete CPU. I know Intel and AMD CPUs include voltage/frequency/power-gating dials for selecting a power/performance point, but I don't know if GPU engines make heavy use of these features.

The simplified GPU being integrated with the CPU probably saves some power from not having to go off-chip to talk to each other. This power savings might be what justifies having two distinct devices.