I'm neither an Unreal nor a C++ expert, but I assume that this kind of GC still allows exact control over when the collection happens since it's not language- but a library feature. In this case this should still be better for game engine purposes, since the GC can for example be hidden behind GPU rendering time.

Totally different from C#:sa GC.

The GC in unreal engine is used to know when what entity should be removed from the game. (If you have multiple entities pointing each other.)

The memory allocation still works how it works in C++.