How to co-design a manycore edge accelerator - Benoît Dupont de Dinechin, Kalray

Intelligent systems are everywhere now, but they require plenty of compute power. How can we deliver the performance to enable artificial intelligence (AI) at the edge, where energy efficiency is also a pre-requisite? In this technical keynote from Computing Systems Week Lyon (25-27 October 2021), Kalray's Benoît Dupont de Dinechin explores different kinds of processor and accelerator architecture, and the design choices involved, showing the rationale behind the Kalray MPPA® manycore processor.

He goes on to set out definitions o edge computing and consider the implications involved, such as prcoessing data as close as possible to the source. Intelligent systems for edge computing, furthermore, combine requirements for cyber-physical systems, artificial intelligence and high-performance computing. Benoît looks at Kalray's response, taking the automotive sector as an example. His talk delves into the MPPA® architecture and supporting software, showing how it responds to these needs.

Benoît finished his talk by explaining how public funding has been integral to this private-sector success story, with the European Union-funded Mont-Blanc and EPI projects supporting this work.