NVIDIA CUDA Tutorial 10: Blocking with Shared Memory
In this tute we'll use a technique called blocking to finally fulfill Porky Water's tall order!
Blocking is a technique where blocks of data are copied from global memory to shared memory, threads work on the data in the much faster shared memory. This greatly reduces the amount of traffic on the global memory bus and allows threads to use the much faster shared memory for most of the calculations.
Blocking with shared memory gives us a great speed up here and easily fulfills Porky's boss's request of a 10x speed up. There's some small changes that could allow the code to run a little quicker but if the code had to run much faster a complete change in algorithm would be far more useful than tweaking this brute force one.
On this page of the site you can watch the video online NVIDIA CUDA Tutorial 10: Blocking with Shared Memory with a duration of hours minute second in good quality, which was uploaded by the user Creel 16 December 2013, share the link with friends and acquaintances, this video has already been watched 20,408 times on youtube and it was liked by 211 viewers. Enjoy your viewing!
