Parallel Computing for Data Science

Chapter 6

The Shared-Memory

Paradigm: GPUs

6.1 Overview

The video game market is so lucrative that the industry has developed ever-

faster graphics cards, in order to handle ever-faster and ever-more visually

detailed video games. These actually are parallel processing hardware de-

vices, so around 2003 some people began to wonder if one might use them

for parallel processing of nongraphics applications. Such programming was

called GPGPU, general programming on graphics processing units, later

shortened to simply GPU programming.

Originally this was cumbersome. One needed to figure out clever ways of

mapping one’s application to some kind of graphics problem, ...