NVIDIA and Arm have expanded their collaboration by enabling NVLink Fusion technology across Neoverse AI data center platforms. The integration allows partners building systems with Arm-based processors to connect their chips directly to NVIDIA accelerators through a high-bandwidth coherent interface that eliminates memory bottlenecks, limiting artificial intelligence performance.
The technology uses the AMBA CHI C2C protocol developed by Arm to create seamless data movement between central processing units and graphics processing units at rack scale. Major cloud providers, including AWS, Google, Microsoft, Oracle and Meta already deploy Neoverse cores, with the architecture expected to capture half the hyperscaler market by 2025.
Partners can leverage NVLink Fusion to build customized energy-efficient AI systems while achieving faster integration and improved time to market. The collaboration builds on previous achievements with Grace Hopper and Grace Blackwell platforms that first demonstrated coherent CPU-GPU integration for high-performance computing workloads.
The technology uses the AMBA CHI C2C protocol developed by Arm to create seamless data movement between central processing units and graphics processing units at rack scale. Major cloud providers, including AWS, Google, Microsoft, Oracle and Meta already deploy Neoverse cores, with the architecture expected to capture half the hyperscaler market by 2025.
Partners can leverage NVLink Fusion to build customized energy-efficient AI systems while achieving faster integration and improved time to market. The collaboration builds on previous achievements with Grace Hopper and Grace Blackwell platforms that first demonstrated coherent CPU-GPU integration for high-performance computing workloads.
