NVIDIA and Arm Collaborate to Introduce NVLink Fusion Support on Neoverse Platforms for High-Bandwidth AI Data Center Interconnects

NVIDIA and Arm have unveiled a significant enhancement to their collaboration with the introduction of NVLink Fusion support for Arm’s Neoverse AI data center platforms. This initiative aims to empower the burgeoning demand for AI-driven infrastructure.

Accelerating AI at Scale: NVIDIA NVLink Fusion on Arm Neoverse AI Data Centers

NVIDIA’s Official Announcement: The data center landscape is undergoing a transformative shift driven by artificial intelligence, characterized by a focus on efficiency and performance metrics. Arm’s Neoverse architecture is at the forefront of this change, powering over a billion cores and anticipated to capture 50% of the hyperscaler market by 2025. Major cloud service providers—such as AWS, Google, Microsoft, Oracle, and Meta—are increasingly leveraging Neoverse, signifying its importance in scaling AI technologies.

In response to the skyrocketing demand for AI solutions, Arm has integrated NVIDIA NVLink Fusion, a high-bandwidth and coherent interconnect technology that was first developed with Grace Blackwell. NVLink Fusion facilitates the connection of CPUs, GPUs, and accelerators into a cohesive rack-scale architecture, addressing memory and bandwidth limitations that can hinder AI performance. Complemented by Arm’s AMBA CHI C2C protocol, it guarantees efficient data flow between Arm-based CPUs and various vendor accelerators.

The collaboration between Arm and NVIDIA is establishing a new benchmark for AI infrastructure, allowing ecosystem partners to develop unique, energy-efficient systems that drive innovation in this AI-dominated era.

“Folks building their own ARM CPU, or using an Arm IP, can actually have access to NVLink Fusion, be able to connect that ARM CPU to an Nvidia GPU or to the rest of the NVLink ecosystem, and that’s happening at the racks and scale-up infrastructure, ” stated Buck.

Arm’s Official Release: Two years prior, Arm and NVIDIA accomplished a landmark achievement with the NVIDIA Grace Hopper platform and NVIDIA NVLink, which delivered a coherent CPU-GPU integration that transformed high-performance computing. To sustain this innovative trajectory, the ecosystem requires versatility and choice—qualities that NVLink Fusion provides by enabling partners to seamlessly integrate Arm-based compute with their preferred accelerators using a coherent, high-bandwidth interface.

The robust momentum and ongoing demand for Grace Blackwell serve as catalysts for the comprehensive expansion of NVLink Fusion across the Neoverse ecosystem, equipping partners to engineer distinctive, energy-efficient AI systems tailored to meet the performance and scalability challenges of the AI revolution.

A NVIDIA processor labeled 'B KR UBF427' is centrally placed on a circuit board surrounded by multiple connected heat sinks.

The robust integration of NVLink Fusion is empowering ecosystem partners to eliminate memory and bandwidth constraints that could limit AI system capabilities, thus enhancing overall performance.

Furthermore, Arm is augmenting the Neoverse platform through the latest release of the AMBA CHI C2C protocol, ensuring compatibility with NVIDIA NVLink Fusion. This enhancement allows Neoverse-based System on Chips (SoCs) to facilitate seamless data transfer between Arm CPUs and diverse accelerators. The outcome is a streamlined integration process, expedited time-to-market, enhanced bandwidth for accelerated computing, and added flexibility for partners developing advanced AI solutions.

As the partnership between Arm and NVIDIA evolves, it ushers in new dimensions of co-design and collaboration, enhancing the intelligence per watt and fundamentally shaping the architecture of the AI era.

Source & Images