NVIDIA ConnectX-8 800Gb/s SuperNIC for AI Networking

• AI and GPU cluster networking
• Large-scale HPC computing
• Hyperscale cloud infrastructure
• Generative AI data centers
• XDR InfiniBand fabrics
• 800Gb/s Ethernet deployments
• Accelerated storage networking
• GPU-to-network data paths
• High-performance distributed computing
• Next-generation data-center modernization

• Up to 800Gb/s networking
• Ethernet and InfiniBand support
• PCIe Gen6 connectivity
• XDR InfiniBand support
• Up to 2x400GbE link configuration
• NVIDIA In-Network Computing acceleration
• Direct NIC-GPU datapath support
• RDMA networking support
• RoCE support
• Quality of service and congestion control
• Socket Direct capability on supported configurations
• OCP 3.0 and CEM PCIe x16 form factors
• Support for high-speed AI compute fabrics
• Linux and Windows ecosystem support
• Designed for AI, HPC and hyperscale environments 

• Brand: NVIDIA
• Product Family: ConnectX-8
• Part Number: ConnectX-8-800G
• Product Type: SuperNIC
• Maximum Data Throughput: Up to 800Gb/s
• Network Technologies: Ethernet, InfiniBand
• Ethernet Speed: Up to 400GbE
• InfiniBand Technology: XDR, NDR, HDR
• PCIe Interface: PCIe Gen6
• PCIe Lanes: Up to 48 lanes
• Form Factors: OCP 3.0, CEM PCIe x16
• Port Configuration: Single-port OSFP224 or dual-port 112, depending on configuration
• Networking Features: In-Network Computing, RDMA, QoS, congestion control
• Management/Software: NVIDIA DOCA, WinOF-2, OpenFabrics Windows Distribution
• Supported Operating Systems: Linux, Windows 

WHY CHOOSE THIS PRODUCT?

• ConnectX-8 is a strong choice for enterprises building 800Gb/s AI, HPC or cloud networking fabrics. Its combination of PCIe Gen6, XDR InfiniBand, Ethernet support and In-Network Computing gives buyers more architectural flexibility than lower-speed ConnectX generations. For procurement, verify the exact adapter configuration, server PCIe platform, switch technology and required port type before ordering, as ConnectX-8 is available in multiple configurations.