Eric Petit (Intel)


Hi everyone,

In about two weeks, we will have the pleasure to have Eric Petit from Intel for one of our second-floor-famous Storm Talks! Please check below for more information.

Speaker

Eric Petit

When

Monday, September 18, 2023 at 10:30 am.

Where

room George Boole 2 or online (Zoom link: https://u-bordeaux-fr.zoom.us/j/88246064235?pwd=SkRSQURqejd6SEtnbmJWckNWalhGZz09)

Title

A small step into the AI new deal for computer architecture and programming.

Abstract

Recent years have seen a tremendous number of new research contributions to computer arithmetic, challenging lower precision arithmetic and rounding mode, leaving the IEEE754 standard far behind. Similarly, the landscape of workloads to support in mainframe systems have drastically shifted from HPC to AI with widely different software practices. This all comes from the rise of AI workloads as one of the main drivers in the data center software and architecture design. Keeping on par with the incredibly fast changing usage of performance libraries, computer arithmetic, compilers and runtimes has pushed our algorithms, tools, and capacity to their limit. On the other hand, the latest AI breakthrough also provides its own opportunities to solve these design problems by leveraging ML algorithms. This is an opportunity to rethink and redesign our approach to hardware and software, promoting new tools and methodologies, and allowing ground breaking solutions to be promoted to mainstream software and hardware implementation. In this talk I will provide some more context about this change and discuss some of the exciting work I am sharing with collaborators inside and outside intel.

Bio

Eric Petit joined Intel in 2016, he is now part of the Math Library Pathfinding group in Hillsboro. He received his Ph.D. from University of Rennes at Inria on compiler technology for GPGPU in 2009. He spent 2 years at University of Perpignan and 6 years at University of Versailles where he led a team participating in multiple European projects on performance analysis tools and parallel runtimes and programming, with multiple open-source software contributions and peer-reviewed publications. His current research interests are on computer arithmetic, and AI for code generation and optimization.

Please note that Eric will stay the whole day at Inria: feel free to chat with him afterwards :)