91³Ô¹ÏÍø

.

AI/ML workloads are pushing the limits of hyperscale data centers and high-performance computing (HPC) systems. Even those that have adopted 400G Ethernet are struggling to connect thousands of graphics processing units (GPUs), support more traffic and higher bandwidth accelerators, and scale at the pace of these modern workloads. To address these challenges and drive the pervasive intelligence of tomorrow, 800G technologies and 1.6T standards are being developed today.

The IEEE P802.3df task force recently released version 1.0 of the 800GbE specification, while the IEEE802.3dj task force continues to work on the 1.6T standard. Collective efforts like the Ethernet Alliance are bringing these standards to life with new, interoperable 800Gbps Ethernet links for rack-to-rack and chip-to-chip applications. The links include electrical backplanes, copper DACs, and electro-optical interfaces, all of which are essential for facilitating the compute, memory, and storage disaggregation required by next-generation AI/ML workloads.

Synopsys joined the Ethernet Alliance at the 2024 European Conference on Optical Communication (ECOC) to showcase the robustness and interoperability of 800G, demonstrating new high-speed interconnects and the progression of standards ¡ª from paper to field deployment.

800G Ethernet Standard in Action

To demonstrate the capabilities of 800G Ethernet at ECOC 2024, the Ethernet Alliance established a with best-in-class switches, routers, and interconnect solutions. The setup included OSFP, QSFP-DD, and QSFP modules in addition to SFP pluggable form factors, traffic generators, and sophisticated test and measurement solutions.

The demonstration featured Auto-Negotiation and Link Training (AN/LT) at 800Gbps, including seamless interoperability, error correction, performance optimization, and reduced manual configuration.

The demonstration helped validate the end-to-end performance and interoperability of silicon-proven 800G Ethernet technology ¡ª across various scenarios and operational metrics, including link testing, latency, and error handling.