Mellanox MQM9700-NS2R

High-performance computing and AI clusters demand a network fabric that can keep pace with massive data flows, and the Mellanox MQM9700-NS2R Quantum-2 InfiniBand switch delivers exactly that. With 64 ports of NDR200 (200 Gb/s per port) and a staggering 25.6 Tb/s of switching capacity, this 1U rack-mount switch eliminates bottlenecks and accelerates time-to-insight for the most demanding workloads.

Who is it for

Engineered for HPC, AI/ML, and cloud data centers, the MQM9700-NS2R is ideal for organizations running large-scale simulations, deep learning training, or distributed storage. It suits system architects and IT teams who require ultra-low latency, high bandwidth, and robust fabric management without compromising on density or power efficiency.

Main features

• 64 QSFP112 ports supporting NDR200 (200 Gb/s each), delivering 25.6 Tb/s aggregate throughput
• Port-to-port latency under 130 ns for real-time responsiveness in parallel computing
• Advanced SHARPv3 in-network computing, adaptive routing, and congestion control for optimized fabric performance
• Dual hot-swappable redundant AC power supplies and front-to-rear cooling for high availability and easy maintenance
• Comprehensive management via CLI, Web UI, SNMP, and REST API, with NVIDIA UFM support for unified fabric orchestration
• Compact 1U form factor saves valuable rack space while maximizing port density

Specifications

• Switch Type: InfiniBand Switch (Quantum-2 generation)
• Port Speed: NDR200 (200 Gb/s per port)
• Number of Ports: 64
• Port Interface: QSFP112
• Switching Capacity: 25.6 Tb/s
• Port-to-Port Latency: <130 ns
• Form Factor: 1U Rack Mount
• Power Supplies: 2x Hot-swappable, Redundant AC
• Cooling: Front-to-Rear Airflow
• Management Ports: 1x RJ45 Ethernet, 1x USB, 1x Serial (RJ45)
• Management Protocols: CLI, Web UI, SNMP, REST API
• Max Power Consumption: 900W (Typical)
• Fabric Management Support: NVIDIA UFM (Unified Fabric Manager)

Tips

To maximize performance, pair this switch with NVIDIA Quantum-2 ConnectX-7 adapters and use NVIDIA UFM for proactive fabric monitoring and optimization. Ensure proper front-to-rear airflow in your rack to maintain optimal operating temperatures. For redundancy, connect both power supplies to separate circuits.