I know there are a lot of gimmicks where you can get outstanding performance if your workload fits within a cache. Feels like maybe we're getting the same sort

I know there are a lot of gimmicks where you can get outstanding performance if your workload fits within a cache. Feels like maybe we're getting the same sort of trickery with thermals. I've seen a lot of over-the-top benchmarks lately from low-power devices, but if you tell them to run the task in a loop for half an hour, they saturate their heatsinks, and drop the clock to the sustained "non-turbo" clock or even throttle below it, so the score on the 10th run is a fraction of what it was on the 1st. On the one hand, this is obviously gaming the knowledge that many benchmarks are short and bursty, but it makes me wonder if there's merit in trying to design a thermal solution that would allow permanent boost clock for thin-and-light-is-irrelevant tasks. If you had a iOS based kiosk or whatever, I could imagine installing a "shucked" iPad that was mounted to a much bigger cooler.