Memory Scalability in AI Co-Design Using CXL Technology

Explore a technical conference talk examining how CXL technology enables new memory architectures for handling massive AI datasets. Learn about the challenges of data movement in AI training and inferencing at terabyte/petabyte scales, focusing on performance and energy costs. Discover how CXL technology creates a paradigm shift by enabling memory disaggregation and hardware coherent memory semantics through PCIe-compliant serial interfaces. Understand the implications of headless NUMA nodes and how this new scalable memory subsystem architecture moves beyond traditional CPU-attached memory. Presented by Rambus experts Danny Moore and Taekang Song, delve into heterogeneous and disaggregated memory approaches while examining potential future applications of this transformative technology.