Coalesced Memory in CUDA

As I always say in my CUDA lectures, taking care of accessing global memory in CUDA is always needed. In this article, Mark Harris presents an study about the consequences of misaligned access in this kind of memory. As is possible to see, absent from stride access, newest GPU architecture are becoming less sensitive to coalesced access. A must read for anyone that wants to learn about how to get more performance in CUDA.

Leave a Reply

Your email address will not be published. Required fields are marked *