Lou Ternullo, Senior Director of IP Product Management at Rambus, discusses the critical role of PCIe and CXL interconnects in enabling the AI-driven data centers of tomorrow. He explores the challenges of traditional server setups, such as inefficiencies in resource allocation, over-provisioning, and data replication, and presents disaggregation as a solution to improve cost-efficiency and performance. By leveraging PCIe and CXL interconnects, data centers can enable heterogeneous compute, memory sharing, and lower-latency communication across servers and racks. Lou underscores the importance of these interconnect technologies in reducing latency, optimizing resource use, and lowering costs in AI-focused data centers.
Why PCIe and CXL are Critical Interconnects for the AI Era
The Road Ahead for Main Memory in the Data Center
Why Memory Matters for AI
The Impact of AI 2.0 on Memory & Interconnect Technology
Addressing AI’s Insatiable Demand For Power
The growth of AI has been staggering, and applications are emerging across the industry that offer new generative AI capabilities powered by large language models. The impact of these AI 2.0 applications is broad and fundamentally alters the way we interact with computers. However, this improvement in capabilities and performance has also been accompanied by massive increases in power consumption.
HBM4 Controller Product Brief
The Rambus HBM4 Controller is designed to support customers with deploying a new generation of HBM memory for cutting-edge AI accelerators, graphics and high-performance computing (HPC) applications. The Rambus HBM4 Controller supports a data rate of 10 Gigabits per second (Gbps) and delivers a total memory bandwidth of 2,560 Gigabytes per second (GB/s) or 2.56 Terabytes per second (TB/s) of throughput for every attached HBM4 memory device.

