In my previous response I said that the Unigine Valley benchmark didn't throttle. That turns out to be incorrect. I ran the benchmark a second time with Ultra settings at 1710x1112 and ran powermetrics (via
asitop) and could see my 10 GPU M2 thermal throttle. The asitop tool shows when the SoC starts to throttle. Peak package power was 18.5W with GPU peak at 9.9W and CPU peak at 6.2W. After the M2 started to throttle package power dropped continuously until it hit about 9W then seemed to level off. The GPU power dropped down to about 3.5W over the same period.
Interestingly the temperatures peaked about 90 °C and then dropped over time to about 70 °C and settled around there. So the throttling isn't temperature based or it doesn't seem that way. I'm using my own temperature monitor app so it might have bugs and maybe isn't showing the peak temperature (oddly the temperature sensors of the M2 are named completely differently than the M1.)
So there is pretty severe GPU throttling with 3D animation over time. It took over 3 minutes before I saw any throttling but then the throttle indicator stayed on until I stopped the benchmark. I don't have an 8 GPU core M2 MacBook Air to test on so I don't know what the characteristics of that MBA look like over the same time period. But the benchmark is free and available and powermetrics is built in to macOS so anyone with an 8 GPU core M2 could compare.
Unigine Valley Benchmark
asitop