Low power consumption Ethernet interoperable link to scale AMD adaptive SoCs for AI/ML
RANOVUS Inc has announced interoperability of AMD Versal adaptive SoCs with the co-packaged Odin 800G direct-drive optical engine and third-party 800G DR8+ retimed pluggable modules.
Odin is a low latency, high density, protocol agnostic, and standards-based optical engine that provides huge optical interconnect bandwidth with industry-leading cost and power efficiency. Constructed on GlobalFoundries Fotonix monolithic RF/CMOS SiPh platform, it includes the company’s proprietary RF CMOS, silicon photonics, laser, and advanced packaging technologies for volume manufacturing. Odin is ideal for next-generation data centre architectures built on co-packaged optics, near-packaged optics, and pluggable OSFP/QSFP-DD/OSFP XD optical modules.
“We announced the first generation of our Odin optical interconnect at OFC 2022 for proprietary AI/ML applications. We are thrilled to showcase our standards-based Odin optical interconnect product with 5pJ/bit for a direct-drive CPO solution,” said Dr Christoph Schulien, head of Systems and High-Speed IC R&D of RANOVUS. “Its inherent versatility enables hyperscale data centre providers to drastically reduce power consumption and optimise density and cost as they deploy novel hybrid data centre architectures in response to the insatiable growth in AI/ML workloads.”
“RANOVUS’ demonstration of interoperability between our Versal adaptive SoCs co-packaged with Odin 800G direct-drive CPO 2.0 and third-party 800G DR8+ retimed pluggable modules underlines the flexibility and scalability of RANOVUS’ technology,” said Yohan Frans, vice president, Engineering at AMD. “We are proud of our collaboration with RANOVUS in demonstrating the performance and versatility of monolithic silicon photonics interconnects as data centre, and 5G customers deploy highly efficient and cost-effective systems for next-generation workloads.”
“RANOVUS’ demonstration of interoperability between CPO and pluggable modules is a key proof point that their interconnect technology supports the flexibility and scalability with the lowest power consumption sought by hyperscalers as they optimise their data centres for AI/ML workloads,” said Vladimir Kozlov, founder and CEO of LIGHTCOUNTING.