NVIDIA InfiniBand Switches

NVIDIA InfiniBand provides AI developers and scientific researchers with the highest networking performance to take on the world’s most challenging problems. Next data rate (NDR) 400Gb/s InfiniBand with new NVIDIA In-Network Computing acceleration engines provides ultra-low latency while delivering the scalability and feature-rich capabilities required for supercomputers, artificial intelligence, and hyperscale cloud data centers.

NVIDIA provides the world’s fastest and smartest switches, enabling in-network computing through NVIDIA Scalable Hierarchical Aggregation and Reduction Protocol (SHARP)™ technology. The NVIDIA Quantum QM8700 series has the highest fabric performance available in the market with up to 16Tb/s of non-blocking bandwidth with sub 130 nanosecond (ns) port-to-port latency. Built with the NVIDIA Quantum InfiniBand switch device, the QM8700 series provides up to forty 200Gb/s full bi-directional bandwidth per port. Additionally, the QM8700 switch is available in both managed and externally managed models.

Based on the NVIDIA Switch-IB^® 2 device, the NVIDIA SB7800 switch series provides up to 36 ports of 100Gb/s full bi-directional bandwidth per port. The SB7800 switch systems are smart network switches that enable in-network computing through NVIDIA SHARP technology. This architecture enables all active data center devices to accelerate the communications frameworks using embedded hardware, resulting in order-of-magnitude application performance improvements. The SB7800 switch is available in both managed and externally managed models.

The NVIDIA SB7780 InfiniBand router enables a new level of scalability and isolation that’s critical for the next generation of data centers. The SB7780 InfiniBand router is based on the Switch-IB application-specific integrated circuit (ASIC) and offers fully flexible 36 enhanced data rate (EDR) 100Gb/s ports, which can be split among six different subnets. The SB7780 InfiniBand router can connect between different types of topologies. Therefore, it enables each subnet topology to best fit and maximize its application’s performance. For example, the storage subnets may use a Fat-Tree topology while the compute subnets may use 3D-torus, DragonFly+, Fat-Tree, or other topologies that best fit the local application. The SB7780 can also help split the cluster to segregate between applications that run best on localized resources and between applications that need a full fabric.

The NVIDIA CS8500 switch series provides up to 800 ports of high data rate (HDR) 200Gb/s InfiniBand, enabling the fastest interconnect speed for HPC and cloud infrastructures. As the world’s smartest network switch, CS8500 switches deliver an optimally performing fabric solution in a 29U form factor, with 320Tb/s of full bi-directional bandwidth and ultra-low port latency. The CS8500 combines the advantages of NVIDIA’s SHARP-based in-network computing, adaptive routing, congestion control, and more to enable order-of-magnitude HPC and AI application performance improvements and extremely high scalability. The CS8500 also leverages NVIDIA’s SHIELD™ (Self-Healing Interconnect Enhancement for Intelligent Data Centers) to overcome link failures, enabling network recovery 5000X faster than any other software-based solution.

The NVIDIA CS7500 series of smart director switches provide up to 648 ports of EDR 100Gb/s InfiniBand—enabling a high-performing fabric solution for HPC environments in an up to 28U form factor. Networks built on the CS7500 series deliver 130Tb/s of full bi-directional bandwidth with 400ns port latency and carry converged traffic combining assured bandwidth and granular quality of service. CS7500 smart network switches also enable in-network computing using NVIDIA SHARP software—an architecture that accelerates the communications frameworks using embedded hardware, resulting in order-of-magnitude application performance improvements. The SB7800 switch is available in three different port configurations: 648, 324, and 216 ports.