Thanks for the article.
Basically, the three things I try to optimize are: Physics objects, Mesh distortion and For each loops. All this works fantastically when there is an optimal number of objects, and when there are a lot of them, then there are minor problems in cpu and gpu %, but it's all solvable.
Now my plan is to reduce the number of physical objects, divide the objects with mesh distortion into groups that will never work at the same time, and to rewiew each part of the project, and all the above instructions, how I can improve the for each loops.
For example, in the case of a large number of physics objects, the real-time management of stepping iterations, if the algorithm of the project allows it, can reduce cpu up to about 10%.