Introduction

The NVIDIA NVLink™ switch system is a rack mount switch that delivers an unprecedented 25.6 Tera Bits per second (Tbps) of full duplex bandwidth for Fourth Generation NVLink in a 1U standard chassis design. Each NVLink Switch includes two Third Generation NVIDIA NVSwitch™ chips and exposes 128 NVLink4 ports via 32 OSFP cages. Each NVLink4 port consists of 2 lanes (fixed width) with Per-link Raw BW of 26.5625GB/s. The third generation of the NVSwitch builds on the advanced communication capability of NVLink to deliver higher bandwidth and reduced latency for compute-intensive workloads. Each NVSwitch has 64 NVLink ports equipped with engines for NVIDIA Scalable Hierarchical Aggregation Reduction Protocol (SHARP)™ for in-network reductions and multicast acceleration.

Increasing compute demands in AI and high-performance computing (HPC) — including an emerging class of trillion-parameter models — are driving a need for multi-node, multi-GPU systems with seamless, high-speed communication between all GPUs. To build the most powerful, end-to-end computing platform that can meet the speed of business, a fast, scalable interconnect with low protocol level overhead is needed. NVIDIA NVLink is purpose designed to match GPU operational characteristics. At 100 Gbps per lane, the Fourth generation NVLink is more than three times faster than PCIe Gen5.

With the NVSwitch, NVLink connections can be extended across NVIDIA DGX™ and NVIDIA HGX™ nodes. By supporting external Fourth Generation NVLink, NVLink connectivity is extended to create a seamless, high-bandwidth, low latency, multi-node GPU cluster — Effectively forming a data center-sized GPU. By adding a second tier of NVLink switches externally to the servers, the NVLink switch system can interconnect up to 256 GPUs.

NVLink Switch Systems Front View
NVLink Switch Systems Rear View

• For a thorough overview of the function of each icon, please see Interfaces and LED Notifications.

• For additional airflow options, see Airflow.

*Each OSFP port contains 4 NVLink4 ports.

The table below describes maximum throughput and interface speed per system model.

*Each OSFP port contains 4 NVLink4 ports.

The table below lists the various management interfaces and available replacement parts per system model.

• For a full feature list, please refer to the system’s datasheet. Go to https://www.nvidia.com/TBD/.

• Additional 3D illustrations can be provided by NVIDIA per-demand.

The list of certifications (such as EMC, Safety and others) per system for different regions of the world is located on the NVIDIA website at https://www.nvidia.com/en-us/networking/environmental-and-regulatory-compliance/.