A fast introduction to GPU acceleration for computing

Description

Graphics processing units (GPUs) used to be mere video cards and have become fully-fledged computing devices no longer connected to visual peripherals. General-purpose GPUs contain hundreds of processors capable of crunching numbers concurrently. The arrangements of memory units and processors, also said architectures, have been evolving continuously to offer higher and higher energy and computing performance.

However, architectures pose constraints on implementing parallelism efficiently. Also, not all algorithms are equally suited for unlimited scalability, whereby the possibility to harvest from parallelism can be constrained from the outset. This not-so-technical talk will offer some orienting knowledge to show where the potency of GPUs lies, and present some notions useful to find your way in a rapidly evolving technological niche.

To this end, being aware of a continuum of situations linking hardware features, software coding requirements, and achievable improvements in computing performance is key. The talk aims to stimulate a sound intuition of what the so-called ´hardware acceleration' with GPUs is and invite reasoned interest in it.

Period	13 Jun 2019
Held at	Fluid Mechanics
Degree of Recognition	Local