NVLink – NVidia Link is a high speed, wire-based, serial, multi-lane, near-range communications link for Nvidia’s GPUs.It has a mesh architecture with no central hub. An endpoint device can have multiple NVLinks. It enables GPUs to communicate directly with each other, bypassing a CPU (x86) host, and offers faster data transfer speeds and lower latency than traditional PCIe-based connectivity. A single Nvidia Blackwell GPU supports up to 18 NVLink 100 gigabyte-per-second (GB/s) connections for a total bandwidth of 1.8 terabytes per second (TB/s ) (900 GB/s in each direction)—2X more bandwidth than the previous generation and over 14X the bandwidth of PCIe Gen5 (16GB/sec per lane).
NVLink Switch chips connect multiple NVLinks to provide all-to-all GPU communication at full NVLink speed within a single rack (up to 72 GPUs per rack) and between racks. To enable high-speed, collective operations, each NVLink Switch has engines for NVIDIA Scalable Hierarchical Aggregation and Reduction Protocol (SHARP) for in-network reductions and multicast acceleration.
Up until May 2025 this interconnect fabric has been limited to Nvidia GPUs and CPUs. The introduction of NVLink Fusion enables x86 CPUs and non-Nvidia-designed accelerators to use NVLink. NVLink Fusion is offered in two configurations;
- Connecting custom CPUs to Nvidia GPUs.
- Using NVLink to connect Nvidia Grace and, in the future, Vera CPUs to non-Nvidia accelerators. This can either be achieved by integrating the NVLink IP into a custom design or via an interconnect chiplet packaged alongside a supported XPU.
