NVIDIA has announced today a new series of network switches, the X800 series, designed for large-scale artificial intelligence (AI) applications. These are the world’s first network platforms capable of delivering end-to-end data transmission speeds of 800 Gb/s, marking a new milestone in network performance for computing and AI workloads.
The new NVIDIA Quantum-X800 InfiniBand and NVIDIA Spectrum™-X800 Ethernet switches are optimized to maximize efficiency in data centers incorporating the new product line based on the NVIDIA Blackwell architecture. “The NVIDIA network is fundamental to the scalability of our AI supercomputing infrastructure,” said Gilad Shainer, Senior Vice President of Networking at NVIDIA. “The X800 switches are end-to-end network platforms that allow us to achieve AI generation at a scale of trillions of parameters, essential for the new AI infrastructures.”
Among the first to adopt the Quantum InfiniBand and Spectrum-X Ethernet solutions are Microsoft Azure and Oracle Cloud Infrastructure. “AI is a powerful tool for transforming data into knowledge. Behind this transformation is the evolution of data centers into high-performance AI engines with a growing demand for network infrastructure,” said Nidhi Chappell, Vice President of AI Infrastructure at Microsoft Azure.
The Quantum-X800 platform sets a new standard in delivering maximum performance for AI-dedicated infrastructure. It includes the NVIDIA Quantum Q3400 switch and the NVIDIA ConnectX®-8 SuperNIC™, which together achieve industry-leading end-to-end transmission speeds of 800Gb/s, representing a 5x increase in bandwidth capacity and a 9x increase in network calculations of 14.4Tflops with Scalable Hierarchical Aggregation and Reduction Protocol (SHARPv4) compared to the previous generation.
On the other hand, the Spectrum-X800 platform delivers optimized network performance for AI cloud and enterprise infrastructure. Utilizing the 800Gb/s Spectrum SN5600 switch and the NVIDIA BlueField®-3 SuperNIC, the Spectrum-X800 platform offers advanced feature sets crucial for multi-tenant generative AI clouds and large enterprises.
NVIDIA also provides a complete set of network acceleration libraries, software development kits, and management software to optimize performance for trillion-parameter AI models. This includes the NVIDIA Collective Communications Library (NCCL), which extends GPU parallel computing tasks to the Quantum-X800 network fabric, leveraging its powerful network computation capabilities with SHARPv4 supporting FP8, boosting performance for large model training and generative AI.
The Quantum-X800 and Spectrum-X800 switches will be available next year through a wide range of leading infrastructure and system providers worldwide, including Aivres, DDN, Dell Technologies, Eviden, Hitachi Vantara, Hewlett Packard Enterprise, Lenovo, Supermicro, and VAST Data, marking a significant advancement in network infrastructure for the next generation of computing and AI.