I know that Go, Lisp and many other compiled languages have Garbage Collectors. Why would they choose to, given the performance cost?
In highly complex, long running systems, it's very difficult to know automatically when memory is no longer needed.
Reference counting, and its draconian creepy uncle malloc()/free(), require programmers to be very careful about when they free memory.
This is a life of tossing in turning in bed: Did you carefully free before you returned in that error handler?
C++ has the benefit of strict compile time analysis, and a lot of the confusing bits have sensibly been pushed to the standard container types' implementations.
Dynamic systems with diverse "types" and millions of collections of objects, especially ones that are likely to contain "cycles" and references to each other, often take other approaches that aren't as "match up the alloc() and free()"-oriented, such as periodically scanning the entire object graph and deleting stuff that couldn't be reached.
Madness! But runs like hell when the graph is small and most people won't feel the burn. An ongoing area of research.