Zero to 50 ExaFLOPS in Under a Year - Lessons from the Trenches

Explore the remarkable journey of scaling AI inference from zero to 50 exaFLOPS in under a year through this keynote presentation by Hagay Lupesko, SVP of AI Inference at Cerebras Systems. Learn about the Cerebras Wafer Scale Engine, the world's largest chip that is 50 times bigger than a GPU and capable of over 100 petaFLOPS of AI compute, and discover how clusters of these engines power Cerebras Inference to deliver the world's fastest AI inference service for industry leaders including Meta, Mistral, and IBM. Gain insights into the critical lessons learned during this massive scaling journey, including understanding what systems break and when, why system design is more crucial than model optimizations, and how to build robust observability, security frameworks, and organizational culture that can survive and thrive at exascale operations. Receive unvarnished, practical insights from the trenches of scaling AI inference infrastructure without having to experience the challenges firsthand.