Why can page-locked Memory be acc in memcpy funciton

Do anybody knows as topic.
page-locked and paged memory are both on host memory.
why can only page-locked be faster when call memcpy? does it already be mapped to device?

there are numerous posts on this subject already.

to copy memory to the device, the device does a DMA on page-locked memory. So pageable memory is first copied to a page-locked buffer before it is transferred by the GPU.
On Core i7 with it’s much faster memory subsystem (and PCI-Ev2 is still the same speed), the difference between page-locked and pageable memory is much smaller.