NVLink Fusion is NVIDIA's high-bandwidth, low-latency connective technology and IP that enables hyperscalers and AI natives to deploy custom XPUs and CPUs into NVIDIA's leading AI infrastructure. It allows mixing heterogeneous compute, sharing racks, networking, cooling and power, and accelerating time-to-market.
The technology comprises NVLink 6 (connecting up to 72 accelerators all-to-all at 3.6 TB/s per XPU, with a roadmap to 1,152 units), NVLink-C2C (extending NVLink to coherent chip-to-chip connections) and NVHBM (a custom high-bandwidth memory architecture). The ecosystem involves CPU, custom-silicon, technology and optical-interconnect partners; Amazon's Annapurna Labs is the first to work with NVIDIA on NVHBM.
High-speed interconnect · serves as: Connectivity / data interconnect.
Which group NVLink Fusion belongs to and how it is built
A subcategory covering interconnects and rack-scale fabrics that link accelerators and processors into coherent, high-bandwidth systems.
A component type covering high-speed interconnects — fabrics and IP providing coherent, high-bandwidth, low-latency GPU–GPU and CPU/accelerator communication, e.g. NVLink.
