I understand the benefits of using pinned host memory for doing copies to and from the device. My problem is that I am capturing an image, copying that to the GPU and processing, then copying back to the host to be viewed. Is the only way I can use the benefits of pinned memory by allocating pinned memory for the source image and copying that data to that pinned memory space? In this case the memcopy to the pinned space would take longer than the whole transfer unpinned transfer to the GPU. The same applies for reading the data from the GPU, would I have to read it to pinned memory and then copy that to a pagable memory space to be read by my view program?
I hope what I’m asking makes sense :huh: