>But the TDP doesn’t have anything to do with how much power the cpu will draw or how much heat it will generate in practice. Those numbers vary quite a lot; the CPU continuously adjusts it’s own frequency multiplier based on it’s own measured temperature, meaning it’ll draw more power if you cool it better.
I don't get it, are you referring to the phenomenon that different workloads have different power consumption (eg. a bunch of AVX512 floating point operations vs a bunch of NOPs), therefore TDP is totally made up? I agree that there's a lot of factors that impact power usage, and CPUs aren't like a space heater where if you let it run at full blast it'll always consume the TDP specified, but that doesn't mean TDP numbers are made up. They still vaguely approximate power usage under some synthetic test conditions, or at the very least is vaguely correlated to some limit of the CPU (eg. PPT limit on AMD platforms).
