Faster unicores are still needed

Abstract

Summary form only given. For the last decade, the advent of the multicore era has driven most of the research architecture effort from the high end processor industry as well as from the computer architecture research community. However, many of the workloads executed on our servers, PCs, smartphones, tablets are still inherently sequential or multiprogrammed; even parallel applications require high sequential performance. Therefore, there are new opportunities for research in uniprocessor microarchitecture. In this presentation, I will first present the motivations of the ERC grant DAL Defying Amdahl's Law. I will then present two research actions on processor core architecture that are on-going within the DAL framework in the INRIA/IRISA ALF group: revisiting value prediction and out-of-order execution of predicated instruction sets.