Heard a lot about how interleaving the instances is usually better performance because of memory access issues, but how exactly does that occur? Is it due to cache properties?

interleaving instances could increase cache hits because one of the instance of the gang would trigger memory request that would then result in loading data into caches that subsequent instances can read directly from.

So the benefit to doing it in another way is that loading the Nth item will cause the cache to accidentally also fetch N+1, N+2, etc?

If memory is loaded from a certain address, why does the "contiguousness" of certain array instances impact the performance? Are these kinds of things already optimized for behind-the-scenes, or does the programmer explicitly have to give details for how they desire the gang of program instances distribute the array among themselves?