Keynote - Multi-Million Core Clusters Co-designed for the Future of Generative Al

Sean Lie, Cerebras, USA; ABSTRACT: Wafer Scale Technology enables multi-million core AI compute clusters to work on generative AI problems. We will show how these clusters eliminate the traditional challenges of distributed compute and how they simply and easily achieve near perfect linear scaling for AI work.  We will also share recent findings showing Cerebra’s techniques for training to state of the art accuracy with few FLOPS and for less power using unstructured sparsity. Finally, case studies will be presented showing clusters of Cerebras CS-2s producing pioneering results in generative AI across industries from Energy to Life Sciences to National Security.