Marvell Technology, Inc. announced industry adoption of the Marvell®Alaska® P PCIe retimer product line, which scales connections between AI accelerators, GPUs, XPUs, CPUs, SSDs, CXL devices and other components inside advanced data center systems. Marvell PCIe retimers enable low-power, high-speed and low-latency connectivity that scales within servers and clusters to power accelerated AI data center infrastructure. Server vendors have adopted Marvell Alaska P PCIe 6 retimers for their GPU and XPU platforms and have also deployed retimer cards in their general-purpose servers. In addition, PCIe active electrical cable (AEC) and active optical cable (AOC) offerings with Marvell Alaska P retimers are available from multiple cable and optical module partners, and the technology has also been evaluated for integration into several storage systems to improve the signal quality between the CPU and SSDs.

Operating at 64 gigatransfers per second (GT/s) using PAM4 signaling, PCIe 6 links require retimers to maintain signal integrity over the increasing distances between components in large-scale, accelerator-rich systems. Built on Marvell 5nm PAM4 SerDes technology, the Marvell Alaska P PCIe retimers family delivers low power and high performance, which compensates for the 40 dB of channel loss that occurs while providing advanced telemetry, diagnostics and fleet-management capabilities. These features allow cloud operators to design efficient, reliable compute fabrics for accelerated infrastructure in the AI era. Developed to support a wide range of different cloud customer data center architectures, Marvell Alaska P PCIe retIMers are purposely built to be used on-board, integrated into copper cable connections or combined with electrical-to-optical components to produce optical PCIe cables.

By connecting both XPU peripherals and scale-up fabrics across AI clusters, Marvell Alaska P PCIe Retimers are at the forefront of accelerated infrastructure, delivering the efficiency, scalability and signal integrity required for AI and cloud compute fabrics.