<< Click to Display Table of Contents >> Navigation:  GPU Processing > Theory of operation > Using the GPU Processing Tool

 

When using the GPU to process images we should always keep in mind that the speed improvement is partly paid with the penalty for transferring the data to the GPU and back to the host for post-processing.
As a consequence we should always try to keep the data in the GPU memory as long as possible to reduce the bottleneck of data transfer.
It might make sense to review your algorithms and apply small changes that allow you to reduce the data to be transferred back and forth.

In a first example this document describes a very simple application where the image data is transferred to the GPU, processed using the GPU and transferred back to the host.
A second example will discuss a pipeline approach that applies different shader programs to the image data.