NVIDIA Spectrum-X Ethernet Networking Platform
Accelerate AI network performance by 1.6x over off-the-shelf (OTS) Ethernet.
Introduction
The World’s First Ethernet Networking Platform for AI
Train and deploy massive-scale AI with the NVIDIA Spectrum-X™ Ethernet networking platform. Powered by the tight coupling of the NVIDIA Spectrum-X Ethernet switch and the NVIDIA Spectrum-X Ethernet SuperNIC™, the platform delivers the highest performance for AI—connecting AI compute fabrics within the data center, scaling across multiple AI data centers with Spectrum-XGS Ethernet, and supporting innovative co-packaged optical switching for enhanced power efficiency and AI workload resiliency.
Benefits
Powerful Performance in the AI Cloud
Spectrum-X Ethernet is a revolutionary solution for building multi-tenant, hyperscale AI clouds. Using Spectrum-X Ethernet, organizations can significantly improve the performance and power efficiency of AI clouds and gain higher predictability and consistency, resulting in faster time to market and a greater competitive edge.
Nearly Perfect Bandwidth at Scale
Extremely Low Latency
End-to-End Stack Optimization
Advanced RoCE Extensions for Scalable AI Communications
Deterministic Performance and Performance Isolation
Open Network Operating System: SONiC and NVIDIA® Cumulus®
Improved AI Cloud Performance
NVIDIA Spectrum-X Ethernet enhances network performance by 1.6x, accelerating the processing, analysis, and execution of AI workloads and, in turn, the development and deployment of AI solutions.
Accelerated Ethernet for Hyperscale
Powered by NVIDIA networking innovations, Spectrum-X Ethernet is fully standards-based Ethernet with support for open Ethernet stacks (SONiC) at cloud scale.
Full-Stack Optimization and Simulation
Spectrum-X Ethernet is tuned and validated across the full stack of NVIDIA hardware and software, creating an unmatched Ethernet solution for AI clouds. NVIDIA DSX Air integrates simulation of Spectrum-X Ethernet with the full-stack AI factory, including best-of-breed components from NVIDIA and ecosystem partners.
Connect Data Centers for Gigascale AI
NVIDIA Spectrum-XGS Ethernet is built for scaling across multiple disparate data centers—located in different buildings or separated by hundreds of kilometers—allowing them to function as a single unified AI super-factory. Spectrum-XGS Ethernet uses a topology-aware congestion control algorithm, precise latency management, and end-to-end telemetry to deliver 1.9x higher NCCL performance in cross-data center environments.
NVIDIA Silicon Photonics
Spectrum-X Ethernet Photonics reduces power consumption, increases reliability, and enhances AI productivity by situating the optics on the same package as the switch ASIC. Spectrum-X Ethernet Photonics enables 5x better network power efficiency, 10x higher network resiliency, and 5x longer sustained AI application runtime compared with traditional pluggable transceiver-based networks.
Spectrum-X Multiplane
Spectrum-X multiplane enables Ethernet to scale from thousands to hundreds of thousands of GPUs using a flat, two-tier topology. By splitting each GPU’s SuperNIC across two or more independent network planes, it overcomes single-plane limits, improves performance and resiliency, and scales Spectrum-X Ethernet networks up to 128K GPUs in two tiers—64x more than single-plane networks.
Rack-Scale Spectrum-X Ethernet
Spectrum-X Ethernet is fully integrated into the NVIDIA MGX™ ETL rack-scale architecture, delivering high-performance Ethernet networking for ultra-dense compute racks and expanding MGX liquid-cooling and reliability innovations to a wider variety of compute options. The power-efficient, copper-based cable cartridge connects liquid-cooled Spectrum-X Ethernet switch trays to up to 256 GPUs in a single rack, while front-panel optical connectivity enables scale-out to larger GPU domains.
Products
What’s Inside the Platform
NVIDIA Spectrum-X Ethernet Switches
The NVIDIA Spectrum-X Ethernet switches are purpose-built to accelerate hyperscale generative AI fabrics. Spectrum-X Ethernet switches deliver high effective bandwidth and performance isolation for AI factories, scaling infrastructure to hundreds of thousands of GPUs and enabling the training and deployment of the world’s largest AI models.
NVIDIA Spectrum-X Ethernet SuperNICs
The NVIDIA Spectrum-X Ethernet SuperNICs are novel classes of network accelerators, purpose-built for supercharging hyperscale AI workloads. Designed for network-intensive, massively parallel computations, NVIDIA SuperNICs provide high-bandwidth RDMA over Converged Ethernet (RoCE) network connectivity between GPU servers, optimizing peak AI workload efficiency.
Applications
How Spectrum-X Ethernet Technology Is Used
Partners
Who We’re Working With
-
Server
-
Storage
-
Orchestration
Resources
Continue Exploring Spectrum-X Ethernet
