AI has come a long way in terms of delivering real value to commercial organizations, and yet there’s still progress to be made before the technology becomes truly pervasive, impacting our world and delivering value everywhere it possibly can. One of the roadblocks to this evolution is the expansion of products and architectures introducing heterogeneity that engineering organizations must master to stay competitive. In this talk, AMD will chart a roadmap for pervasive AI across data center environments, beginning with our introduction of the ZenDNN library, tightly integrated with popular AI frameworks so developers can enable inference on EPYC processors. We’ll cover the current features of ZenDNN enabling a broad arrange of apps to be accelerated, outlining the use cases already available today. From there, we’ll explain how AMD’s Unified Software Stack ties into ZenDNN and AMD EPYC CPUs. Lastly, we’ll share how our vision of heterogenous deployment of inference enablement across all architectures, with a single software stack so developers can more easily build, optimize, and deploy AI solutions for real world use cases.