CyberTAN unveils liquid-cooled AI networking platform using Marvell silicon

CyberTAN Technology has introduced the Pallas Series, a liquid-cooled 102.4 Terabits per second (Tbps) Ethernet switch platform designed to address thermal and power challenges in AI clusters. Developed in collaboration with Marvell, the architecture is engineered to support Marvell’s Teralynx T100 102.4 Tbps switch silicon.

The Pallas Series utilizes proprietary Direct-to-Chip (D2C) liquid cooling technology capable of managing over 1,000 watts of localized heat dissipation. This design aims to enable high port density and improve Power Usage Effectiveness (PUE) for hyperscale data centers managing generative AI workloads. The platform is built to accommodate the 3nm Marvell T100 architecture, leveraging its energy-saving capabilities.

While the announcement marks an engineering milestone at the OCP Global Summit, the current Pallas platform serves as a prototype using a thermal module to simulate chip workloads. No specific customer orders or shipment volumes have been disclosed at this stage.