"Fundamentally, it’s a function that maps a slice of GPU memory into the host’s address space, typically for streaming geometry data or texture uploads"

Can someone clarify this for me? Are OpenGL/D3D buffers that get stuff memcpy'd into them by the CPU actually "slices of GPU memory," or are they more often reserved driver memory that eventually get DMA'd to the GPU? (I realize both probably happen at different times, but I'm curious which is more typical for modern systems)

It seems like spending CPU cycles writing every byte over the bus would perform much worse than a fast write to sysmem followed by a DMA transfer.

EDIT: I looked into it, and it seems like the typical implementation is that map returns a pointer to some pinned driver sysmem and unmap kicks off an async DMA to GPU memory.