As hyperscale data centre infrastructures evolve to support the processing of trillions of parameters in large language models, they need to scale to hundreds of thousands of accelerators with highly efficient and fast connections.
Synopsys Ultra Ethernet and UALink IP will provide a holistic, low-risk solution for high-speed and low-latency communication to scale-up and scale-out AI architectures.
Commenting Neeraj Paliwal, senior vice president of IP product management at Synopsys said, "With the industry's first Ultra Ethernet and UALink IP, companies can get a head start on developing a new generation of high-performance chips and systems with broad interoperability to scale future AI and HPC infrastructure."
The advanced capabilities of Synopsys’ Ultra Ethernet IP solution include:
IP solution to scale out backend networks: Consisting of PHY, MAC and PCS controller, and verification IP, the Synopsys Ultra Ethernet IP solution offers a low-risk path for designers developing systems that can support up to one million endpoints in a single network.
Best-in-class 224G Ethernet PHY IP: The silicon-proven Synopsys 224G Ethernet PHY IP, which supports the Ultra Ethernet protocol, has demonstrated broad interoperability.
Patented error correction implementation: The Synopsys Ultra Ethernet MAC and PCS controller IP provides up to 1.6 Tbps of bandwidth with ultra-low latency, enabling the real-time processing needed for AI workloads.
Seamless integration: The MAC and PCS IP support an interface to the higher layers of the Ultra Ethernet stack providing a full silicon implementation for switches, AI accelerators and smart NICs.
Accelerated verification and validation: The Synopsys Ultra Ethernet verification IP helps ensure protocol adherence to rapidly evolving standards, enabling faster and more efficient validation of AI and HPC systems.
"Juniper has already introduced the industry's first 800GbE capability with its PTX10002-36QDD Packet Transport Router, which utilises our proprietary Express 5 ASIC with Synopsys Ethernet IP," said Debashis Basu, senior vice president of Juniper Engineering. "We will continue to partner with Synopsys and leverage the latest technologies from the Ultra Ethernet Consortium (UEC) to transition into the 1.6TbE era.
“This indicates our ongoing innovation in high-speed networking to achieve our goal to significantly improve scale, reliability, and performance in data centre networks. This is particularly important as AI workloads continue to grow exponentially, making such networks much more efficient and cost-effective."
The advanced capabilities of the new Synopsys UALink IP solution include:
IP solution to scale up computing fabrics: Consisting of PHY, controller, and verification IP, the Synopsys UALink IP solution speeds time-to-market for designers developing systems that can support up to 1,024 AI accelerators.
Efficient, high-speed data transfers: Engineered for data-intensive AI workloads, the low-power and high-bandwidth Synopsys UALink PHY IP provides 200 Gbps per lane.
Latency-optimised with memory sharing capabilities: Synopsys UALink Controller IP helps mitigate critical bottlenecks of AI hardware infrastructure via shared memory access from accelerator to accelerator.
Built-in protocol checks: The Synopsys UALink Verification IP, combined with Synopsys hardware-assisted verification solutions, provides quick and reliable verification for AI hardware.
"Synopsys has decades of expertise in contributing to the industry's essential interconnect standards and in delivering widely adopted high-speed interface IP," said Kurtis Bowman, chairperson of the board at UALink Consortium. "We appreciate Synopsys' commitment to enabling UALink IP to create a scalable, high-performance data centre ecosystem for designers to meet the growing demands of AI models."
The Synopsys Ultra Ethernet IP solution, including MAC and PCS, PHY, and verification IP, is scheduled to be available in the first half of 2025.
The Synopsys UALink IP solution, including controller, PHY, and verification IP, is scheduled to be available in the second half of 2025.
