NVIDIA launches 800Gb/s networking platforms for AI
© Reuters.
SAN JOSE, Calif. – NVIDIA (NASDAQ:) has unveiled its latest networking platforms, the Quantum-X800 InfiniBand and Spectrum-X800 Ethernet, which are the first in the industry to deliver end-to-end 800Gb/s throughput, targeting the burgeoning demand for high-performance AI infrastructure.
The announcement was made today, with the new platforms designed to enhance computing and AI workloads across various data centers.
The Quantum-X800 series includes the Quantum (NASDAQ:) Q3400 switch and the ConnectX-8 SuperNIC, boasting a bandwidth capacity five times higher than its predecessors and a ninefold increase in In-Network Computing to 14.4Tflops.
This significant advancement is facilitated by NVIDIA’s Scalable Hierarchical Aggregation and Reduction Protocol (SHARPv4), which is tailored to support large-scale AI applications.
On the other hand, the Spectrum-X800 series is optimized for AI cloud and enterprise infrastructure, featuring the Spectrum SN5600 800Gb/s switch and the NVIDIA BlueField-3 SuperNIC. This series is specifically designed for multi-tenant generative AI clouds and large enterprises, ensuring performance isolation for each tenant’s AI workloads.
According to Gilad Shainer, senior vice president of Networking at NVIDIA, the X800 switches are pivotal for achieving trillion-parameter-scale generative AI, which is crucial for new AI infrastructures. Early adopters of the new networking solutions include tech giants such as Microsoft (NASDAQ:) Azure and Oracle (NYSE:) Cloud Infrastructure, with Coreweave also joining the ranks.
NVIDIA’s software support extends to a suite of network acceleration libraries and software development kits, enhancing the performance of AI models with trillion parameters.
The Collective Communications Library (NCCL) leverages GPU parallel computing tasks across the Quantum-X800 network fabric, utilizing SHARPv4 to support FP8 and significantly boosting performance for large model training and generative AI.
The Quantum-X800 and Spectrum-X800 platforms are expected to be available next year from a variety of leading infrastructure and system vendors globally. NVIDIA, known for pioneering accelerated computing since its inception in 1993, continues to expand its full-stack computing infrastructure offerings that are transforming various industries.
The information for this article is based on a press release statement from NVIDIA.
This article was generated with the support of AI and reviewed by an editor. For more information see our T&C.
