NVIDIA Spectrum-X networking platform
The First Ethernet Networking Platform for AI in the World
With the NVIDIA Spectrum-X networking platform, advanced AI applications can be provisioned and managed with ease. With the help of the NVIDIA BlueField-3 SuperNIC and the NVIDIA Spectrum-4 Ethernet switch, Spectrum-X offers the best performance for a variety of industry applications as well as AI, machine learning, and natural language processing.
NVIDIA Boosts Generative AI via Ethernet Networking
The Spectrum-X Cloud service providers, GPU cloud providers, and enterprises have adopted NVIDIA Networking; its wide system-maker support expands its market reach.
Among the first AI cloud service providers that use NVIDIA Spectrum-X to deliver exceptional networking performance to their AI infrastructures are CoreWeave, GMO Internet Group, Lambda, Scaleway, STPX Global, and Yotta. Furthermore, a number of NVIDIA partners have announced Spectrum-based products, joining Dell Technologies, Hewlett Packard Enterprise, Lenovo, and Supermicro in integrating the platform into their products. These partners include ASRock Rack, ASUS, GIGABYTE, Ingrasys, Inventec, Pegatron, QCT, Wistron, and Wiwynn.
NVIDIA Spectrum-X completely transforms Ethernet networking, enabling companies to fully utilize their AI infrastructures to revolutionize their operations and industries.
Annual Release Cadence
The first Ethernet fabric designed specifically for artificial intelligence, Spectrum-X, improves generative AI network performance by 1.6 times compared to conventional Ethernet fabrics.
NVIDIA founder and CEO Jensen Huang announced that the company intends to introduce new NVIDIA Spectrum-X products annually, offering more bandwidth and ports as well as improved software feature sets and programmability to drive leading AI Ethernet networking performance. This is in response to the industry’s strong demand for the performance that Spectrum offers.
Spectrum-X is an end-to-end platform designed with the performance and functionality needed by generative AI clouds. It includes the NVIDIA Spectrum SN5600 Ethernet switch and the NVIDIA BlueField -3 SuperNIC. The highest-performance Ethernet networking for AI is provided by utilizing adaptive routing and congestion control to maximize bandwidth and isolate noise, yielding consistent results for thousands of concurrent AI workloads at all sizes.
The platform provides zero-trust security, GPU computing elasticity, and enhanced cloud multi-tenancy when paired with NVIDIA BlueField-3 DPUs. It allows cloud service providers to increase their return on investment while speeding up the creation and implementation of AI solutions.
Key Features and Benefits
NVIDIA Spectrum-X is a cutting-edge approach of using Ethernet to create multi-tenant, hyperscale AI clouds. By using it, businesses may greatly increase the AI clouds’ performance and power efficiency as well as their predictability and consistency, which will give them a competitive advantage and a quicker time to market.
- Almost Ideal Bandwidth at Scale
- Exceptionally Low Latency
- Optimizing the entire stack
- Advanced RoCE Extensions for AI Communications That Are Scalable
- Performance Isolation and Deterministic Performance
- Cumulus and SONiC are open network operating systems.
Improved AI Cloud Performance
By increasing network performance by 1.6X, NVIDIA Spectrum-X speeds up AI workload processing, analysis, and execution, which in turn speeds up the creation and implementation of AI solutions.
Accelerated Ethernet for Hyperscale
Spectrum-X, a fully standards-based Ethernet that supports open Ethernet stacks (SONiC) at the cloud scale, is driven by NVIDIA networking advancements.
Full-Stack Optimization
With its full stack of NVIDIA hardware and software tuning and validation, Spectrum-X offers an unparalleled Ethernet solution for AI clouds.
Spectrum-4 from NVIDIA
The fifth generation of Spectrum Ethernet switches, the NVIDIA Spectrum SN5000 series, are designed specifically to speed up hyperscale generative AI fabrics. SN5000 switches provide accelerated Ethernet for any data center without sacrificing features or performance, with port rates up to 800 gigabits per second (Gb/s).
NVIDIA BlueField-3
An innovative family of network accelerators designed specifically to boost hyperscale AI workloads is the NVIDIA BlueField-3 SuperNIC. The BlueField-3 SuperNIC maximizes the efficiency of AI workloads by offering up to 400Gb/s RoCE network connectivity across GPU servers, which is ideal for network-intensive, massively parallel computations.
