which one is faster? Texture or Page-Locked Host Memory

[CUDAkk]

Hi All,

I have read about Page-Locked Host Memory that

Zero copy is a new feature as of version 2.2 of the CUDA Toolkit. It enables GPU threads to directly access host memory. For this purpose, it requires mapped pinned (nonpageable) memory. On integrated GPUs, mapped pinned memory is always a performance gain because it avoids superfluous copies as integrated GPU and CPU memory are physically the same.

and about texture memory

The read-only texture memory space is cached. Therefore, a texture fetch costs one device memory read only on a cache miss; otherwise, it just costs one read from the texture cache.

What I observed in some extent texture is fast and some other extent Page-Locked(pinned) Host Memory excess is fast.

Then I am querious to know in Texture & Page-Locked Host Memory which one is most efficient to use?

[jma]

You can’t use a texture as a replacement for neither pinned nor zero-copy.

You would instead use a texture as a cache on top of something you have copied (or zero-copied) to the device memory space.