Enquire: Asia & Africa - +65-98008081 USA - +1-919-995-4114

Home Networking HPE NVIDIA MQM9790-NS2F 400Gb/s NDR InfiniBand Switch

HPE NVIDIA MQM9790-NS2F 400Gb/s NDR InfiniBand Switch

The HPE P45691-B21 / NVIDIA MQM9790-NS2F is a 1U Quantum-2-based NDR InfiniBand switch designed for high-density AI, HPC, and data-center fabrics. It provides 64 NDR ports at 400Gb/s through 32 OSFP cages, with unmanaged/external fabric management and power-to-connector (P2C) forward airflow. Its 51.2 Tb/s bidirectional switching capacity makes it suitable for scalable GPU and high-performance compute interconnects.

OVERVIEW

  • The MQM9790-NS2F is a 1U NVIDIA Quantum-2 NDR InfiniBand switch for dense AI/HPC fabrics, offering 64 × 400Gb/s NDR ports via 32 OSFP cages and 51.2Tb/s bidirectional throughput. Compatibility: designed for NDR InfiniBand fabrics and corresponding OSFP-based infrastructure. Model comparison: MQM9790-NS2F is unmanaged with P2C airflow; MQM9700-NS2F is the managed counterpart. 

USE CASES

  • AI GPU training clusters - Provides high-density 400G NDR connectivity for distributed GPU workloads where fabric bandwidth can become a major scaling constraint.
  • High-performance computing (HPC) - Creates a high-throughput InfiniBand fabric for compute-intensive scientific and engineering workloads.
  • Large-scale AI infrastructure - 64 NDR ports in a 1U chassis help increase fabric port density while conserving valuable rack space.
  • GPU-to-GPU and server-to-server fabric connectivity - 400Gb/s NDR ports provide the high-speed interconnect required by modern accelerated computing environments.
  • AI cluster spine/fabric deployments - High aggregate switching capacity supports scalable fabric architectures as additional compute nodes are introduced.
  • Data-intensive research environments - Suitable for environments moving large datasets between compute resources where high network throughput is commercially important.
  • Externally managed InfiniBand fabrics - The unmanaged MQM9790 architecture fits deployments where fabric management is provided externally rather than through an internally managed switch model.
  • Data-center racks requiring P2C airflow - Forward power-to-connector airflow can be selected when the rack's thermal design calls for this airflow direction.

KEY FEATURES

  • 64 × 400Gb/s NDR InfiniBand ports
  • 32 OSFP cages
  • 51.2Tb/s bidirectional throughput
  • NVIDIA Quantum-2 architecture
  • Unmanaged switch configuration
  • P2C forward airflow
  • 1U rack form factor
  • High-speed NDR InfiniBand connectivity
  • Designed for AI and HPC environments
  • High-density GPU cluster networking
  • Suitable for scalable data center fabrics
  • HPE OEM configuration

TECHNICAL SPECIFICATIONS

  • Brand: NVIDIA Mellanox
  • Product Family: NVIDIA Quantum-2 QM9790
  • System Model: QM9790
  • Legacy OPN / Model: MQM9790-NS2F
  • HPE SKU: P45691-B21
  • NVIDIA SKU: 920-9B210-00FN-0D0 per current NVIDIA documentation
  • Product Type: NDR InfiniBand switch
  • InfiniBand Generation: NDR
  • Port Count: 64 NDR ports
  • Port Speed: Up to 400Gb/s per NDR port
  • Physical Connector Cages: 32 OSFP
  • Aggregate Bidirectional Throughput: Up to 51.2Tb/s
  • Unidirectional Switching Capacity: 25.6Tb/s
  • Management: Unmanaged / externally managed
  • Airflow: Power-to-connector (P2C), forward
  • Form Factor: 1U rack-mount
  • Architecture: NVIDIA Quantum-2
  • Deployment Compatibility: NDR InfiniBand fabric; verify OSFP transceivers/cables and host adapters for the intended topology
  • Lifecycle Status: Mass Production
  • Alternative Managed Model: MQM9700-NS2F - 64-port NDR, 32 OSFP, managed, P2C
  • Alternative Reverse-Airflow Model: MQM9790-NS2R - 64-port NDR, 32 OSFP, unmanaged, C2P
  • Related Managed Reverse-Airflow Model: MQM9700-NS2R - 64-port NDR, 32 OSFP, managed, C2P

Why Choose This Product?

  • High-density 400G connectivity - 64 NDR ports provide substantial connectivity in a compact 1U chassis.
  • Built for AI and HPC fabrics - NDR InfiniBand is suited to infrastructure where distributed compute requires high-bandwidth interconnects.
  • 51.2Tb/s aggregate capacity - Provides significant switching capacity for large-scale east-west traffic between compute nodes.
  • Space-efficient deployment - 1U form factor helps maximize rack utilization as GPU and compute clusters grow.
  • External-management architecture - A strong fit for organizations whose InfiniBand fabric management is handled through an external management strategy.
  • Correct airflow for P2C racks - The NS2F variant can be selected specifically for power-to-connector forward-airflow environments.
  • HPE OEM procurement path - P45691-B21 gives HPE customers a recognizable OEM SKU for sourcing and infrastructure standardization.
  • Scalable Quantum-2 platform - Provides a high-performance networking foundation for expanding AI/HPC environments rather than requiring a low-bandwidth networking refresh as cluster demands increase. 

CUSTOMER REVIEW

No Reviews Found

FAQ

No FAQs Found