Is there an analogue of Moore's law, but on nuclei?

Question

Multithreading

Hello everyone, I'm used to constantly trying to find a place to use multithreading / concurrency in my code, but lately this has begun to seem not very necessary, even some kind of option by and large. After all, if the complexity of the algorithm is O(n^2) or O(C^n) (yes, even O(n)), then what's the difference how many times it will be faster due to parallelism? Another thing is if there are a lot of cores or parallel threads. Well, let's say that I have hardware with 2000 potential cores (there are 2k potential parallelized threads on them, and this is still a constant, but not bad enough to take into account the complexity for some n), the only question is whether all 2k threads will work ? Here there are suspicions already with 16 threads (when the speed increases not by ~16 times, but less), perhaps the OS or the scheduler or the buses in the system cannot cope with such a number of threads (cannot control them), is there such a problem and how can it be solved? Is there an analogue of Moore for cores so that we can confidently say that with a huge number of cores, the speed of the parallelized algorithm will decrease by an order of magnitude?

Answer 1

Amdahl's law