NVIDIA ConnectX-7 NDR 400G InfiniBand Adapter Card, MCX75310AAS-NEAT

NVIDIA ConnectX-7 NDR 400G InfiniBand Adapter Card

Accelerate Data-Driven Scientific Computing with In-Network Computing

The NVIDIA® ConnectX-7 NDR 400G InfiniBand Adapter Card, part number MCX75310AAS-NEAT, delivers high-speed networking for high performance computing (HPC), artificial intelligence (AI), scientific research, and hyperscale cloud environments. With up to 400 Gb/s throughput and ultra-low latency, this standup half-height, half-length (HHHL) adapter is designed to support demanding compute and data-intensive workloads while helping data centers scale efficiently.

ConnectX-7 integrates NVIDIA In-Network Computing engines that accelerate collective operations, MPI Tag Matching, All-to-All processing, and programmable datapath functions. Remote direct-memory access (RDMA), send/receive semantics, hardware-based congestion control, and atomic operations help optimize communication between systems and improve application performance. Compatibility with the InfiniBand Trade Association (IBTA) Specification 1.5 supports interoperability and long-term infrastructure planning.

The NVIDIA ConnectX-7 platform supports advanced InfiniBand capabilities, including up to four ports, 16 million input/output channels, eight virtual lanes plus VL15, and maximum transmission unit settings from 256 bytes to 4 kilobytes. It also supports messages up to 2 GB. The adapter uses an octal small form-factor pluggable (OSFP) connector and a PCIe standup design, making it suitable for enterprise servers, HPC clusters, AI infrastructure, and other performance-focused data center deployments.

Features:

• Product Type: 400 Gigabit Ethernet card
• Form Factor: Standup HHHL design
• Network Technology: NDR InfiniBand with throughput up to 400 Gb/s
• Connector: Octal small form-factor pluggable (OSFP)
• Compliance: InfiniBand Trade Association (IBTA) Specification 1.5 compliant
• Port Support: Up to four ports
• Memory Access: Remote direct-memory access (RDMA) with send/receive semantics
• Traffic Management: Hardware-based congestion control
• Processing: Atomic operations and NVIDIA In-Network Computing acceleration
• Input/Output: 16 million input/output channels
• Message Support: Messages up to 2 GB
• Transmission Unit: Supports settings from 256 bytes to 4 kilobytes
• Virtual Lanes: Eight virtual lanes plus VL15