How to make your code run faster using GPUs (Alejandro Estana / LPT / Seminar). – 12/04, 14 H.

Alejandro Estana (LPT)

Abstract :

If the sequential execution of our code is not fast enough for our purpose, we can start thinking about parallelizing it. Parallelize a code means that some parts of the code will be run in parallel and therefore the total computing time will be reduced. Because of its parallel processing architecture, the Graphics Processing Units (GPU) has become an essential component when we think about parallelising a code. New clusters contains GPUs in their computing nodes and if we want to take advantage of their computational power we need to adapt our code one way or another.

In the first part of the talk, I will introduce the GPU architecture and GPU programming explaining the different approaches that we can take to use the GPUs power.

In the second part of the talk, I will present three practical examples of GPU parallelization that I have implemented in collaboration with researchers of the LPT and IRAP (Institut de recherche en astrophysique et planétologie).