The NVIDIA Spectrum Ethernet Platform
Explore innovative, efficient Ethernet solutions for accelerated AI.
Introduction
Complete Ethernet Solutions, From Host to Switch
Experience industry-leading Ethernet performance, availability, and ease of use across a wide range of applications—including cloud computing, data storage, AI, and more. NVIDIA Spectrum™ Ethernet is a standards-based, end-to-end platform that includes Spectrum switches, NVIDIA ConnectX™ SuperNIC™s, NVIDIA BlueField™ DPUs, LinkX™ cables, and networking software.
The NVIDIA Spectrum-X™ Ethernet platform is purpose-built to drive generative AI at unprecedented scale—boosting network performance by 1.6x and sustaining breakthrough 95% efficiency across deployments exceeding 100,000 GPUs.
New Co-Packaged Silicon Photonic Networking Switches to Scale to Millions of GPUs, Multi-Site AI Factories
Products
What’s Inside the Spectrum Ethernet Platform
Spectrum Ethernet Switch Systems
The NVIDIA Spectrum Ethernet switch family includes a switch and software portfolio spanning 1GbE to 800GbE. NVIDIA Spectrum Ethernet switches are ideal for building AI fabrics connecting NVIDIA AI factories, supporting both traditional pluggable optics as well as co-packaged silicon photonics-based switches. Combining Spectrum switches with the NVIDIA Cumulus™ Linux network operating system (NOS), NVIDIA Pure SONiC NOS, NVIDIA Minerva™ AIOps platform, and NVIDIA DSX™ Air simulation platform delivers the performance and features that every data center needs.
BlueField DPUs
The NVIDIA BlueField DPU ignites unprecedented innovation for modern data centers, delivering a broad range of advanced networking, storage, and security services for complex compute and AI workloads, virtual machines (VMs), containers, and bare-metal servers in enterprise and cloud. By combining the industry-leading NVIDIA ConnectX network adapter with an array of Arm® cores, BlueField offers purpose-built hardware acceleration engines with full programmability through NVIDIA DOCA™.
Ethernet SuperNICs
NVIDIA Ethernet SuperNICs are a novel class of network accelerators based on the BlueField-3 networking platform, purpose-built for supercharging hyperscale AI workloads. Designed for network-intensive, massively parallel computing, the ConnectX and BlueField SuperNICs provide up to 1.6 Tb/s RoCE network connectivity between GPU servers, optimizing peak AI workload efficiency. Ushering in a new era of AI cloud computing, the NVIDIA SuperNICs enable high-performance, multi-tenant data center environments while ensuring deterministic and isolated performance between tenant jobs.
ConnectX NICs
The industry-leading NVIDIA ConnectX family of intelligent network adapters offers the broadest and most advanced hardware acceleration engines. Running any workload at any scale, ConnectX enables the highest ROI and lowest TCO for cloud data centers, enterprise environments, and SDN and network functions virtualization (NFV) for telecommunications companies.
Networking Software
NVIDIA networking delivers cloud-scale efficiency and simplified, unified operations via an API-first, automation-centric approach. From low-level SDKs and libraries (DOCA) to switch network operating systems (Cumulus Linux, SONiC), to powerful DOCA services, intelligent AIOps and full-stack observability (NVIDIA Minerva), and simulation (DSX Air) toolsets, NVIDIA’s offerings deliver the features needed for AI factories.
LinkX Ethernet Cables and Transceivers
NVIDIA LinkX cables and transceivers are designed to maximize the performance of high-performance computing networks, delivering high bandwidth, low latency, and highly reliable connections between Ethernet elements.
Capabilities
How Ethernet Enhances the Network
Partners
Spectrum-X Ethernet Networking Platform Available From These Partners
Resources
Discover More About NVIDIA Networking
Next Steps
Ready to Get Started?
Get a Free Trial of NVIDIA Networking Software
Simulate a fully automated network topology using NVIDIA DSX Air.
Discover Networking for the Era of AI
The network is ultimately responsible for AI performance, acting as the backbone of the data center to harness the power of generative AI.
Learn About the Spectrum-X Ethernet Platform
The Spectrum-X Ethernet platform is designed specifically to improve the performance and efficiency of Ethernet-based AI infrastructure.
NVIDIA Spectrum Ethernet FAQs
The NVIDIA Spectrum Ethernet platform is a standards-based, end-to-end platform offering industry-leading performance up to 800Gb/s that includes Spectrum switches, NVIDIA ConnectX SuperNICs, NVIDIA BlueField DPUs, LinkX cables, and networking software. NVIDIA AI factories are connected with Spectrum Ethernet hardware coupled with networking software such as NVIDIA Cumulus Linux network operating system (NOS), NVIDIA Minerva AIOps toolset, and NVIDIA DSX Air simulation platform.
The NVIDIA Spectrum-X Ethernet platform is the world’s first end-to-end, AI-optimized Ethernet platform for multi-tenant generative AI factories. It combines purpose-built Spectrum Ethernet switches (a subset of the switches within the Spectrum Ethernet platform) and SuperNICs to deliver high effective bandwidth, low latency, and strong performance isolation for hyperscale AI infrastructure—all built on standards-based Ethernet.
ConnectX NICs provide broad hardware acceleration for any workload at any scale. ConnectX SuperNICs are a more powerful class of network accelerator purpose-built for hyperscale AI—delivering up to 1.6 Tb/s of RoCE connectivity with adaptive routing and deterministic, isolated performance between tenant jobs. BlueField data processing units (DPUs)—the processor powering the operating system of AI factories—manage and secure the infrastructure, enabling zero-trust tenant isolation, bare-metal compute instances, and full software-defined infrastructure control, powered by NVIDIA DOCA.
NVIDIA LinkX cables and transceivers are designed and validated for the Ethernet Spectrum platform at up to 800Gb/s, where consistent performance is critical. LinkX eliminates the signal integrity issues common with third-party cables, ensuring end-to-end platform performance as tested and certified.
An NVIDIA Ethernet SuperNIC is a purpose-built network accelerator designed specifically for hyperscale AI workloads—distinct from a standard NIC. The NVIDIA ConnectX SuperNIC provides up to 1.6 Tb/s of RoCE network connectivity between GPU servers with hardware-accelerated adaptive routing and Direct Data Placement, delivering deterministic and isolated performance between tenant jobs. Working in co-design with Spectrum-X Ethernet switches, it sustains 95% effective bandwidth across deployments exceeding 100,000 GPUs.
NVIDIA Spectrum-X delivers the high-performance Ethernet fabric—combining ConnectX SuperNICs and Spectrum switches to move data between GPU servers with adaptive routing, deterministic performance, and 95% effective bandwidth at scale. NVIDIA BlueField—the processor powering the operating system of AI factories—complements this by managing and securing the infrastructure, enabling zero-trust tenant isolation, bare-metal compute instances, and full software-defined infrastructure control. Powered by NVIDIA DOCA, BlueField enables enterprises and service providers to scale AI securely and efficiently across the agentic AI factory that Spectrum-X connects.
RoCE (RDMA over Converged Ethernet) enables low-latency data transfers directly between GPU memory over Ethernet, bypassing the CPU. NVIDIA’s Spectrum-X Ethernet implementation of RoCE extensions combine switch telemetry and adapter-side rate limiting in hardware to ensure a lossless fabric and eliminate hotspots proactively—delivering 1.6x higher performance and 95% efficiency for AI workloads versus standard ECN-based approaches.
NVIDIA silicon photonics (co-packaged optics) integrates optical components directly with the Spectrum ASIC, reducing insertion loss from 22 dB to ~4 dB—delivering 63x better signal integrity, 5x better power efficiency, and 2x bandwidth per lane. The SN6810-LD and SN6800-LD (SN6000 series) use MMC-16 connectors, enabling AI factories to scale to millions of GPUs with lower power and simplified cabling.
NVIDIA Minerva is an AI-native AIOps platform that correlates infrastructure and workload telemetry to provide full-stack observability across compute, networking, storage, and AI workloads, and converts these into insights for proactive or automated action. NVIDIA Minerva combines high-frequency telemetry from Spectrum-X Ethernet switches, SuperNICs, NCCL, schedulers, and more to ensure the AI fabric remains healthy, delivering highest performance and maximal uptime for gigascale agentic inference.
NVIDIA DSX Air is a cloud-hosted simulation platform that builds full-stack digital twins of AI data center infrastructure—including networking, compute, storage, and security—before a single piece of hardware is deployed. Teams use DSX Air to design, test, and validate Spectrum Ethernet topologies and automation workflows in a risk-free virtual environment, accelerating deployment and reducing risk.
