Rambus introduces PCIe 7.0 Switch IP with TDM for AI and data center systems

Rambus has introduced a new PCIe 7.0 Switch IP with Time Division Multiplexing, aimed at AI, cloud, and high performance computing systems that need faster and more efficient data movement.

The new switch IP is designed for systems where CPUs, GPUs, accelerators, and NVMe storage must move huge amounts of data with low latency. That problem is becoming more important as AI servers grow larger and more complex.

PCIe 7.0 already brings a major bandwidth increase, but Rambus is focusing on how that bandwidth is used. Time Division Multiplexing helps schedule and share traffic across PCIe links more efficiently. This can help reduce bottlenecks when multiple parts of a system need access to the same high speed fabric.

The company says the new IP is built for next generation AI and data center SoCs. It is designed to support large scale AI training, inference, data movement, and newer system designs that use disaggregated or pooled compute resources.

That means the technology is not aimed at normal desktop PCs. Its main role is in advanced server hardware, where companies need to connect many processors, accelerators, storage devices, and networking components without wasting bandwidth.

Here is a quick look at the key details:

Rambus says the switch IP is part of its wider PCIe 7.0 portfolio, which also includes controllers, retimers, and debug tools. The goal is to help chip designers build faster AI infrastructure while managing performance, power, and reliability needs.

The announcement also shows how AI hardware is moving beyond raw GPU performance. Faster accelerators matter, but they need high speed connections to stay fed with data. Technologies like PCIe 7.0 switching and smarter traffic management are becoming more important as AI systems scale.