NVIDIA and Arm have taken a significant step forward in their collaboration, unveiling NVLink Fusion support for Neoverse AI data center platforms. This move marks a notable advancement in the AI infrastructure landscape, promising enhanced performance and energy efficiency.
NVIDIA NVLink Fusion Integrates with Arm Neoverse for Enhanced AI Capabilities
AI is driving a transformative change in data center architectures, where efficiency per watt has become a critical metric. At the heart of this transformation is Arm Neoverse, which is currently deployed in over a billion cores, with expectations to capture 50% of the hyperscaler market by 2025. Major players like AWS, Google, Microsoft, Oracle, and Meta are already leveraging Neoverse, highlighting its pivotal role in AI scale-up.
To cater to the increasing demands of AI workloads, Arm is expanding the Neoverse platform with NVIDIA NVLink Fusion. This high-bandwidth, coherent interconnect was initially introduced with Grace Blackwell and now links CPUs, GPUs, and accelerators into an integrated rack-scale architecture. By eliminating memory and bandwidth bottlenecks, NVLink Fusion significantly boosts AI performance. It seamlessly connects with Arm’s AMBA CHI C2C protocol, facilitating smooth data movement between Arm-based CPUs and preferred accelerators.
Together, NVIDIA and Arm are setting new benchmarks for AI infrastructure. Their partnership enables ecosystem partners to develop unique, energy-efficient systems that drive AI innovation forward.
“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,” said Buck.
Transformative Impact on AI Infrastructure
Two years ago, Arm and NVIDIA achieved a milestone with the NVIDIA Grace Hopper platform and NVIDIA NVLink, setting a new standard for CPU–GPU integration in high-performance computing. As the ecosystem continues to evolve, flexibility and choice become paramount, and NVLink Fusion offers partners the capability to connect Arm-based compute with their preferred accelerators through a robust, coherent interface.
The expanding demand for Grace Blackwell is now driving NVLink Fusion’s integration across the entire Neoverse ecosystem. This development empowers partners to craft energy-efficient AI systems on Arm, meeting the rigorous performance and scalability needs of the AI era. The collaboration between Arm and NVIDIA is advancing co-design efforts and enhancing intelligence per watt, shaping the future architecture of AI technology.
Ecosystem partners are embracing NVLink Fusion to address memory and bandwidth issues that have constrained AI performance. Designed to interface with AMBA CHI C2C — a technology pioneered by Arm — NVLink Fusion provides a vital protocol for a coherent, high-bandwidth link between CPUs and accelerators.
By building on this foundation, Arm ensures Neoverse compatibility with the latest AMBA CHI C2C protocol, facilitating seamless data transfer between Arm-based CPUs and other accelerators. This leads to faster integration, reduced time to market, higher bandwidth, and increased flexibility for partners developing next-gen AI systems.