Mantle, followed by Vulkan and all, threw out the assumptions about cpu/gpu starving that were made in the 90s to adapt to modern hardware, extracting sometime massive performance boost "simply" by changing the API.
Right now you have RTX IO (nvidia) and DirectStorage (direct x) who aime to extract massive disk IO performance boost from modern NVMe drive for gaming by throwing away assumptions made about them in pre-ssd times and providing a new API.
These change often come down to the same thing: the way we call them, the amount of locking and round trips we do, synchronous or not, the connecting interfaces and their own speed limits, ... Rules made in an era were drive were X orders of magnitude slower than RAM start limiting us when X has been divided by two or three.