Akila Subramaniam

Sr Fellow, AMD

Title: Application and SW-HW codesign approach to driving low power architecture for GenAI

Abstract

AIPCs have ushered in a new era of AI compute that enables inferencing AI models locally versus running them on expensive cloud servers. However, these devices are constrained by cost, power, memory storage and BW. Scaling the compute and memory BW especially for generative AI requires a top-down approach that starts from compilation and low-level transformation of model architectures, low precision datatypes, optimization of the whole pipeline in addition to the HW microarchitecture. We will cover each of these topics and trends in computing and memory that influence the architecture and design of AIPCs.

Bio

Akila Subramaniam is a Senior Fellow leading low-power AI hardware and software initiatives in the Computing and Graphics Business Unit at Advanced Micro Devices (AMD), a global leader in semiconductor and chip design. Her team is responsible for developing hardware and software architectures that enable efficient, low-power AI applications across AI PCs and graphics products. She holds a Master of Business Administration in Management from Lehigh University College Of Business and a MS degree from the University of Cincinnati.