Let's say that I want to efficiently update the position of a bunch of vertices every frame. Normally I would put those vertices inside of a resource backed by an upload heap, then map a cpu pointer to it directly and modify the values through that pointer. From what I have read about the copy queue, it is the fastest way to transfer over PCIe. Could this transfer of data every frame be done better through a copy queue somehow? Or is the speed advantage of copy queues only for CopyResource() type of operations?

Cheers!