As promised last time, in the third and final installment we would look at some actual code. As it will turn out, our straightforward implementation will need a little rework in order to properly fire up the GPU.

Last week we had a quick glance of what GPU programming is and what kind of issues it can tackle. Today, we will dive a bit under the hood and take a look at the logical architecture of modern chips, and show a basic strategy when to migrate part of our code over to the CUDA platform.

Over the last year, there has been quite some fuss over using GPU’s for other tasks than gaming. News stories about AI and cryptocurrency mining requiring vast amounts of GPU’s; product shortages in the consumer gaming market are just a few to name. Share prices for semiconductor design companies, particularly Nvidia, have been soaring. Yet, on the other hand, there is little understanding in the general software development community as to how these techniques work, what class of problems they can solve more effectively, and how and when to use them.