DataCenterNews Asia Pacific - Specialist news for cloud & data center decision-makers
Story image
Nvidia unveils world-first 800Gb/s networking platforms for AI
Tue, 19th Mar 2024

NVIDIA has made waves in the AI and computing industry by launching its X800 series, the world's first networking platforms with end-to-end 800Gb/s throughput. Both the NVIDIA Quantum-X800 InfiniBand and the NVIDIA Spectrum-X800 Ethernet platforms boast software that heightens AI, cloud, data processing, and HPC applications. They are poised to revolutionise data centres of various scales, integrating seamlessly with NVIDIA's Blackwell architecture-based products.

Senior Vice President of Networking at NVIDIA, Gilad Shainer, shared his excitement on the release, saying, "NVIDIA Networking is central to the scalability of our AI supercomputing infrastructure. NVIDIA X800 switches are end-to-end networking platforms that enable us to achieve trillion-parameter-scale generative AI essential for new AI infrastructures." It is clear that the wide-ranging applications of the new switches have been acknowledged by industry heavyweights, with Microsoft Azure and Oracle Cloud Infrastructure being initial adopters of the Quantum InfiniBand and Spectrum-X Ethernet.

Vice President of AI Infrastructure at Microsoft Azure, Nidhi Chappell, explained the transformational power of AI, stating, "AI is a powerful tool to turn data into knowledge. Behind this transformation is the evolution of data centres into high-performance AI engines with increased demands for networking infrastructure." With the integration of NVIDIA's latest networking solutions, Microsoft Azure aims to lead the charge in AI-supported cloud technology.

Significantly, NVIDIA's Quantum-X800 platform sets a stellar industry standard for AI-focused Infrastructure. Paired with the NVIDIA Quantum Q3400 switch and the NVIDIA ConnectX-8 SuperNIC, the platform allows an unparalleled end-to-end throughput of 800Gb/s. It features a staggering five times higher bandwidth capacity compared to the previous generation, and a nine times increase of 14.4 Tflops of In-Network Computing with NVIDIA's Scalable Hierarchical Aggregation and Reduction Protocol (SHARPv4).

A parallel performer, the Spectrum-X800 platform offers optimized networking performance designed for AI cloud and large enterprise infrastructures. In partnership with the Spectrum SN5600 800Gb/s switch and the NVIDIA BlueField-3 SuperNIC, Spectrum-X800 provides advanced features underlining its key role in nurturing generative AI clouds and large enterprises. Importantly, the Spectrum-X800 enables optimal and consistent performance levels for each tenant's AI workload, improving customer satisfaction and the quality of services.

NVIDIA's comprehensive software support ensures the optimised performance of trillion parameter AI models. With software such as the NVIDIA Collective Communications Library (NCCL), the Quantum-X800 network fabric boosts parallel computing, markedly enhancing performance for large model training and generative AI. NVIDIA's full-stack software strategy fosters data centre networks' flexibility, reliability, and responsiveness, thus increasing overall operational efficiency and catering to modern applications' needs.

Infrastructure and system vendors worldwide, including Aivres, DDN, Dell Technologies, Eviden, Hitachi Vantara, Hewlett Packard Enterprise, Lenovo, Supermicro, and VAST Data, are expected to make the Quantum-X800 and Spectrum-X800 available next year. NVIDIA's pioneering product suite is set to reshape the industry, meeting the exponential demand for highly competent and AI-driven networking platforms.