Marvell Launches Custom UALink Scale-Up Solution for AI Infrastructure
Marvell Technology, Inc. a leader in data infrastructure semiconductor solutions, has announced the launch of its custom Ultra Accelerator Link (UALink) scale-up solution. This new offering is designed to enhance performance and scalability in next-generation accelerated infrastructures, addressing the growing needs of AI workloads and hyperscalers striving to expand their AI infrastructure.
Interoperable IP Portfolio for Custom AI Compute Platforms
Marvell’s custom UALink solution offers a high-efficiency, low-latency interconnect for AI platforms, built on open standards. It features 224G SerDes, UALink Controller, scalable Switch Core and Fabric IP, along with advanced packaging options like co-packaged copper and optics.
These components work together to create a highly efficient interconnect system for hundreds or thousands of AI accelerators in scale-up deployments.
With Marvell’s custom silicon capabilities, compute vendors can build tailored solutions, integrating custom accelerators with UALink controllers and switches. This combination, paired with Marvell’s advanced packaging technology, ensures optimal performance for rack-scale AI systems.
Addressing Hyperscaler Challenges in AI Infrastructure
Hyperscalers have long struggled with the need to scale their AI infrastructures while ensuring high performance. Marvell’s custom UALink solution aims to address these challenges by providing a toolkit based on open standards that facilitates direct, low-latency communication between accelerators.
The system supports flexible, scalable switch topologies, empowering hyperscalers to build AI infrastructures capable of supporting the most demanding workloads with the required efficiency and interoperability.
Also read: Marvell Expands AI Infrastructure with Advanced Multi-Die Packaging Solution
Collaboration with AMD and the UALink Consortium
Nick Kucharewski, senior vice president and general manager, Cloud Platform Business Unit at Marvell, said, “We are pleased to introduce our new custom UALink offering to enable the next generation of AI scale-up systems.”
He added, “This addition to our custom portfolio enables customers with flexibility to optimize their AI infrastructure with standards-based scale-up switch and interconnect technology.”
Marvell is also a member of the UALink Consortium, an open industry group focused on developing UALink specifications. This consortium is leading the charge in establishing a connection standard between accelerators and devices, ensuring seamless interoperability and high-performance computing for next-generation AI applications.
