Microchip Technology Inc. is tackling the big bandwidth and super low latency needs of AI and ML with their new XpressConnect™ PCIe 6.0 and CXL 3.1 retimer family. This product line is built to handle serious signal integrity and latency issues in next-gen cloud and enterprise data centers.
As AI accelerators, GPUs, and CPUs move to PCIe 6.0 (64 GT/s) and CXL 3.1, higher frequencies shrink how far data can travel on regular PCB traces before it gets messed up. Microchip’s XpressConnect retimers step in to boost and regen these signals, letting high-speed links in a server chassis stretch further. It does this without adding packet latency or compromising on security either.
Also Read: Edgecore Launches Praxis: A Highly Versatile Edge AI Platform Tailored for AI Service Providers
“AI data centers are increasingly constrained not by compute, but by the ability to move data efficiently across the system. As PCIe 6.0 pushes speeds to 64 GT/s, signal reach and latency become critical design challenges,” said Brian McCarson, corporate vice president and GM of Microchip’s data center solutions business unit. “Our XpressConnect retimers are designed to act as the high performance nerve center of the AI server, helping customers build more scalable, power efficient fabrics by reducing latency and improving connectivity across dense GPU clusters. This system level approach allows data center architects to reclaim underutilized resources and improve overall platform efficiency at scale.”
Key Technical Innovations & Performance Benefits
This tech is super fast, with ultra-low pin-to-pin latency, making sure AI tasks zip right through without hitting bottlenecks. Plus, it’s flexible – supporting both PCIe 6.0 and CXL 3.1 at the same time. This means data centers can easily share memory between different kinds of processors like CPUs, GPUs, and FPGAs.
Also, built-in real-time tools help monitor signal quality and catch issues early, which keeps systems online and running smoothly. The designers thought about energy too; advanced saving modes make it way easier to handle heat in those packed AI servers.
To fit all sorts of setups, it comes in different sizes. You can find it on regular motherboards or special high-density boards for accelerators. So, it really meets a lot of needs out there in data centers.
Elevating AI and CXL Connectivity
Generative AI models need huge computing power, so they require big clusters working together. This means the parts need fast communication. The XpressConnect family with CXL 3.1 helps by letting systems share memory easily. They can shift resources on the fly, cutting costs and improving hardware use.
Microchip‘s new gear solidifies their spot as a major supplier for AI hardware needs. It fits well with their other stuff like PCIe switches and clock managers.
