Since the beginning of the 2010 decade, the use of graphics cards (GPUs) as acceleration devices for certain types of computation has grown rapidly and, while they were initially designed for video games, their use for scientific computation has grown steadily.
Although they offer very significant potential gains in acceleration for certain types of calculation, a notable obstacle to the widespread use of graphics cards was the need for very low-level programming using specific libraries, which required a high degree of computer expertise. But recently, the programming of graphics cards can be done by simple sets of directives (OpenACC standard and OpenMP standard since its version 4.5), in the straight line of the usual programming paradigms on scientific computers, which makes the use of GPUs easily affordable for a large number of programmers.
The objective of this training is to present these new sets of guidelines and to show, through their practical implementation, how they allow today to take advantage of the enormous potential of graphics cards to accelerate scientific applications.