To understand why one large Metal buffer can be worse than multiple overlapping views, you have to look at how the **GPU memory hierarchy**, **CPU-GPU synchronization**, and **driver optimizations** interact. 

In Metal, a `MTLBuffer` is a single, contiguous allocation of memory. While it is the