device access directly hard disk


I am relatively new to cuda so this may be a dummy question.

I know we can allocate sectors in host (cpu) then transfer to gpu and then write back to host, but I was wondering if it possible to bypass the data transfer and read/write disk sectors directly.
Are the msdn functions readfile(), writefile() applied in device code? Can we write something from device mem to hard disk?

Any help appreciated,

CUDA devices can’t access peripherals, like hard disks. The CPU code has to manage disk I/O, and load data into GPU memory.

(Given the rather large speed difference between PCI-Express and SATA, I don’t think there would be any advantage to bypassing the CPU…)