NVIDIA MQM9790-NS2F 400Gb/s InfiniBand Switch – 64-Port NDR, 51.2Tb/s,

The NVIDIA MQM9790-NS2F is a fixed-configuration Quantum-2 InfiniBand switch based on the industry-leading NDR 400Gb/s architecture. Designed for extreme-scale HPC and AI workloads, this externally managed variant provides 64 non-blocking ports (via 32 OSFP connectors) in a compact 1U form factor, delivering an astonishing 51.2Tb/s of bidirectional throughput and over 66.5 billion packets per second.

The MQM9790-NS2F is optimized for environments where a centralized Subnet Manager (such as NVIDIA UFM) orchestrates fabric control. It includes SHARPv3 in-network computing, adaptive routing, RDMA, and advanced congestion control - enabling researchers and AI developers to tackle the most demanding data-intensive challenges with unprecedented performance and scalability.

• 64 ports of 400Gb/s NDR (32 OSFP connectors)
• 51.2Tb/s non-blocking aggregate throughput
• >66.5 billion packets per second (BPPS)
• SHARPv3 in-network computing (32x AI acceleration)
• 1+1 redundant hot-swappable AC power
• Hot-swappable fan units, 80 PLUS Gold+ certified
• Port-split technology: up to 128 ports at 200Gb/s
• Adaptive routing, RDMA, and congestion control
• Backward compatible with HDR and EDR
• RoHS-6 compliant, CE / FCC / VCCI certified

The NVIDIA Quantum-2 platform introduces the third generation of SHARP technology (SHARPv3), delivering up to 32X higher AI acceleration than the previous generation. By offloading collective operations from the CPU to the switch network, SHARPv3 dramatically reduces MPI operation time and data traversal across the fabric.

Quantum-2 ASIC integrates advanced features: port-split for double-density 200Gb/s, adaptive routing, enhanced VL mapping, and self-healing network capabilities. The onboard x86 Coffee Lake i3 CPU runs MLNX-OS for comprehensive management.

SHARPv3 Integration: Supported with MLNX-OS and UFM Enterprise (externally managed).

• Top-of-rack leaf / spine in extreme-scale HPC clusters
• AI training and inference fabrics for deep learning
• High-performance storage and data analytics platforms
• Research & scientific computing (weather, genomics, physics)
• Cloud data centers with large-scale multi-tenant workloads

• Host Adapters: ConnectX-6, ConnectX-7 NDR
• Cables: OSFP passive/active copper, active optical, optical modules
• Management: NVIDIA UFM, MLNX-OS, OpenSM
• Backward: HDR (200Gb/s), EDR (100Gb/s) with appropriate transceivers
• Topologies: Fat Tree, SlimFly, DragonFly+, multi-dimensional Torus

Note: MQM9790-NS2F is the externally managed, P2C airflow variant. Verify your data center cooling strategy before ordering. For internally managed deployments, consider MQM9700 series.

• 1-year standard warranty (extended available)
• Technical support via email / phone
• Firmware upgrades and MLNX-OS updates
• RMA services for defective units
• Global logistics and delivery

Is this switch internally managed?
No, MQM9790-NS2F is externally managed. Requires a host-based Subnet Manager (UFM / OpenSM).

Can I change the airflow direction?
No, P2C (Front-to-Rear) is fixed. Choose -NS2R for C2P airflow.

Does it support SHARPv3?
Yes, with MLNX-OS and UFM Enterprise configuration.

What is the maximum port count at 200Gb/s?
Up to 128 ports using port-split technology.

• Confirm airflow orientation matches your rack cooling strategy.
• Use only NVIDIA-qualified OSFP optics and cables for NDR compliance.
• Ensure Subnet Manager is properly configured before connecting to fabric.
• Operating temperature: 0°C-40°C; maintain adequate ventilation.
• This switch supports 200-240VAC only - not compatible with 100-120VAC.

• ASIC: NVIDIA Quantum-2
• Ports: 64 x NDR (32 OSFP)
• Throughput: 51.2 Tb/s
• Packet Rate: >66.5 BPPS
• SHARPv3: 32x AI acceleration
• Management: External (UFM/OpenSM)

• HCAs: ConnectX-6, ConnectX-7
• Cables: OSFP DAC / AOC / optics
• SM: UFM Enterprise, OpenSM
• Backward: HDR (200G), EDR (100G)
• Topologies: Fat Tree, SlimFly, DragonFly+, Torus