The Pallas Series platform utilizes Direct-to-Chip liquid-cooling to manage over 1,000 Watts of localized heat, addressing the thermal density challenges posed by modern AI clusters. Engineering the system around Marvell’s 3nm T100 silicon allows for high port density while leveraging the chip's inherent 25% energy-saving potential. According to CyberTAN CEO GwongYih Lee, this architecture provides hyperscalers with a production-ready path to low-latency, high-density connectivity that prioritizes strict environmental sustainability.
In section Releases
CyberTAN Debuts 102.4 Tbps Liquid-Cooled Switch Blueprint
At the OCP Global Summit 2026, CyberTAN Technology unveiled its Pallas Series, a 102.4 Tbps Ethernet switch architecture designed to tackle the thermal constraints of AI data centers. By integrating Marvell’s Teralynx T100 silicon with proprietary liquid-cooling, the platform seeks to redefine power efficiency for hyperscale networking.
At the summit, CyberTAN is demonstrating a prototype at Booth B29 that highlights the platform's thermal management capabilities. The live setup features a Coolant Distribution Unit and multi-point sensors to simulate heavy workloads. Tests confirm the system maintains an industry-leading Power Usage Effectiveness of 1.1, even when running 40W OSFP ZR/ZR+ optics under a 45°C liquid inlet temperature. This design aims to resolve traditional networking bottlenecks by balancing massive bandwidth throughput with aggressive thermal control.
Comments (0)
No comments yet. Be the first!