Events
The Evolution of Computer Architecture to Service Machine Learning and Scientific Computing
The talk provides a condensed overview of how computer processor chip architectures evolved from the early days of processors for mainframes, workstations, and desktops to graphics-optimized processors (GPUs), and the latest machine-learning-optimized processors like Amazon Trainium. The talk walks through the first principles driving end-to-end optimization across the compute stack: chip design, instruction sets, compilers and profilers, and how to design systems for an era where most code is written by LLMs.
Speaker Bio
Nafea Bshara, Vice President and Distinguished Engineer at Amazon Annapurna Labs, the silicon arm of Amazon
Nafea Bshara leads the strategy and architecture for Amazon custom hardware like Nitro, Graviton, and Amazon’s Trainium AI chips. These chips build the majority Amazon Web Services compute infrastructure, and run Claude, ChatGPT, and millions of startups, enterprises, education and public sector compute workloads.Nafea began his career at Galileo Technology, where he held various leading roles in software and chip design for network, storage, and compute infrastructure, culminating in his position as ChiefArchitect. Following Galileo’s acquisition by Marvell Semiconductor in 2001, Nafea joined Marvell as CTO for enterprise products.In 2011, he co-founded (together with his friend from Technion days, Hrvoje Bilic) Annapurna Labs, a startup building cloud-optimized infrastructure chips and associated software. Amazon acquired Annapurna Labs in 2015, after which Nafea and his team have ledAmazon’s custom silicon and hardware efforts, and defined new industry categories like DPU.Nafea holds an M.Sc. degree in Electrical and Computer Engineering from the Technion – Israel Institute of Technology. Nafea has been granted over 350 US patents.