NVIDIA Quantum-2 InfiniBand Switch · QM9700 Series

NVIDIA MQM9790-NS2F 400Gb/s InfiniBand Switch, 64 Ports, 32 OSFP, 1U

The MQM9790-NS2F is a 1U, 64-port 400Gb/s InfiniBand switch from the NVIDIA Quantum-2 QM9700 series. It delivers up to 51.2 Tb/s aggregate bidirectional throughput and more than 66.5 billion packets per second, with 32 OSFP connectors, externally managed UFM operation and power-to-connector (P2C) forward airflow.

B2B supply and project configuration support provided by Hong Kong Starsurge Group Co., Limited. Availability, lead time and pricing are confirmed per order.

Product Overview

The NVIDIA MQM9790-NS2F is part of the Quantum-2 QM9700 series of 400Gb/s InfiniBand fixed-configuration switches. It provides 64 ports of 400Gb/s InfiniBand in a 1U standard chassis, with an aggregated bidirectional throughput of 51.2 Tb/s and more than 66.5 billion packets per second of capacity. The system uses 32 octal small form-factor pluggable (OSFP) connectors, supporting passive or active copper and active fiber cabling as well as optical modules.

The platform brings high speed, extremely low latency and scalability to HPC and AI fabrics, incorporating Remote Direct Memory Access (RDMA), adaptive routing and NVIDIA Scalable Hierarchical Aggregation and Reduction Protocol (SHARP). NVIDIA InfiniBand also provides self-healing network capabilities, quality of service (QoS), enhanced virtual lane (VL) mapping and congestion control for high overall application throughput.

As the externally managed variant in the QM9700 family, the MQM9790-NS2F can utilize NVIDIA Unified Fabric Manager (UFM) feature sets for provisioning, monitoring, management, preventive troubleshooting and maintenance of the modern data center fabric. Its P2C (power-to-connector) airflow matches standard front-to-rear rack cooling designs.

Key Features

• 64 × 400Gb/s InfiniBand ports in a 1U non-blocking switch, with aggregate data throughput up to 51.2 Tb/s.
• More than 66.5 billion packets per second capacity for extreme-scale HPC and AI systems.
• 32 OSFP connectors supporting passive or active copper and active fiber cable, plus optical modules.
• NVIDIA SHARPv3 technology for in-network aggregation, with 32X higher AI acceleration power compared to the previous generation.
• RDMA, adaptive routing, QoS, enhanced VL mapping and congestion control for high application throughput.
• Port-split technology enabling a double-density radix for 200Gb/s data speeds, supporting up to 128 ports of 200Gb/s.
• Externally managed via NVIDIA UFM for fabric-level provisioning, monitoring and preventive maintenance.
• 1+1 redundant hot-swappable power supplies, 80 Gold+ and ENERGY STAR certified, 200–240Vac input.
• Hot-swappable fan unit with P2C (power-to-connector) forward airflow in the NS2F model.

Technology

NVIDIA Quantum-2 400Gb/s InfiniBand

As HPC and AI applications become more complex, the demand for advanced high-speed networking is critical for extreme-scale systems. NVIDIA Quantum-2 is positioned as the industry-leading switch platform in power and density, with NDR 400Gb/s InfiniBand throughput. The QM9700 and QM9790 switch systems deliver 64 ports of 400Gb/s InfiniBand per port in a 1U standard chassis, with 51.2 Tb/s aggregate bidirectional throughput and more than 66.5 billion packets per second capacity.

SHARPv3 and In-Network Computing

The QM9700 400Gb/s InfiniBand switches extend NVIDIA In-Network Computing technologies and introduce the third generation of NVIDIA SHARP technology, SHARPv3. SHARPv3 creates virtually unlimited scalability for small and large data aggregation through the network and allows 32X higher AI acceleration power compared to the previous generation. It boosts application performance of complex computations while data moves through the data center network, participating in the application's runtime and reducing the amount of data needed to traverse the network.

Topology Flexibility and Backward Compatibility

As a rack-mounted InfiniBand solution, the QM9700 and QM9790 400Gb/s InfiniBand fixed-configuration switches allow maximum flexibility through a variety of topologies, including Fat Tree, SlimFly, DragonFly+, multi-dimensional Torus and more. They are backwards compatible to previous generations and include expansive software ecosystem support.

Port-Split Technology for 200Gb/s Density

By implementing NVIDIA port-split technology, the QM9700 and QM9790 switches provide a double-density radix for 200Gb/s data speeds, reducing the cost of network design and network topologies. Supporting up to 128 ports of 200Gb/s, NVIDIA delivers the densest top-of-rack (TOR) switch available on the market. The QM9700 family enables small to medium-sized deployments to scale with a two-level Fat Tree topology while reducing power, latency and space requirements.

Management with NVIDIA UFM

The externally managed QM9790 switch can utilize the advanced NVIDIA Unified Fabric Manager (UFM) feature sets to empower data center operators to efficiently provision, monitor, manage, preventatively troubleshoot and maintain the modern data center fabric, realizing higher utilization and reducing overall opex. Note that the internally managed QM9700 model features an on-board subnet manager for out-of-the-box bring-up of up to 2,000 nodes; that on-board subnet manager description applies to the QM9700, not the externally managed QM9790.

Typical Deployments

• Extreme-scale HPC: fabrics that need 400Gb/s per-port throughput and very low latency for complex simulations and highly parallelized algorithms.
• AI training and inference clusters: data-intensive environments that benefit from SHARPv3 in-network computing and RDMA.
• Research and scientific computing: high-resolution simulations and extreme-size datasets requiring real-time information exchange.
• Small to medium-sized deployments: two-level Fat Tree topologies that scale with reduced power, latency and space requirements.
• Dense top-of-rack aggregation: 200Gb/s double-density designs using port-split technology, up to 128 ports of 200Gb/s.
• Government, healthcare, manufacturing, education, finance and enterprise: organizations building scalable, manageable InfiniBand infrastructure.

Compatibility

The MQM9790-NS2F is an NVIDIA Quantum-2 QM9700-series 400Gb/s InfiniBand switch with 64 ports of 400Gb/s InfiniBand and 32 OSFP connectors. It is externally managed through NVIDIA UFM. Compatibility depends on the specific fabric design, cable plant, optical modules and host channel adapters in your environment.

• OSFP connectors supporting passive or active copper and active fiber cable, plus optical modules; exact cable and module compatibility should be confirmed against the NVIDIA compatibility list for your deployment.
• Backwards compatible to previous InfiniBand generations, with expansive software ecosystem support.
• Management ports: 1 × USB 3.0, 1 × USB for I2C channel, 1 × RJ45, 1 × RJ45 (UART).
• Externally managed via NVIDIA UFM; MLNX-OS software package.
• Power input: 200–240Vac; confirm site power and P2C airflow orientation before ordering.

For fabric-level compatibility, optical module qualification and firmware matching, please confirm with our technical team before ordering.

Specification Table

Source: MQM9700 series datasheet provided by the customer. Note: the datasheet ordering description uses “unmanaged” for the QM9790 externally managed model; this page uses “externally managed via UFM” for clarity. Parameters not publicly specified in the source are marked accordingly.

Selection Guide

The QM9700 family includes managed and externally managed 400Gb/s InfiniBand switches with forward and reverse airflow options. Use the table below to match management model and airflow direction to your rack design and fabric role.

Selection note: confirm airflow direction against your rack hot/cold aisle design, and confirm managed versus externally managed operation against your fabric management architecture before ordering.

Advantages

• High density in 1U: 64 × 400Gb/s InfiniBand ports and 51.2 Tb/s aggregate throughput in a standard 1U chassis.
• Extreme-scale performance: more than 66.5 billion packets per second and extremely low latency for complex HPC and AI workloads.
• In-network computing: SHARPv3 delivers 32X higher AI acceleration power compared to the previous generation.
• Topology flexibility: supports Fat Tree, SlimFly, DragonFly+, multi-dimensional Torus and more, with backward compatibility to previous generations.
• Cost-efficient 200Gb/s density: port-split technology provides up to 128 ports of 200Gb/s, reducing network design and topology costs.
• Fabric visibility: externally managed via NVIDIA UFM for provisioning, monitoring, preventive troubleshooting and maintenance.
• Operational efficiency: 1+1 redundant hot-swappable power supplies, 80 Gold+ and ENERGY STAR certified, hot-swappable fan unit and P2C forward airflow option.

Service & Support

Hong Kong Starsurge Group Co., Limited supports this product with an experienced sales and technical team. Support scope can include pre-sales configuration review, airflow and power confirmation, compatibility checks, and order follow-up.

• Pre-sales selection support for port count, airflow, managed versus externally managed operation, and fabric management model.
• Compatibility confirmation for OSFP optical modules, copper and fiber cables, and host adapters before ordering.
• Order configuration and lead-time coordination.
• Multilingual support and global delivery available through Starsurge.

FAQ

What is the MQM9790-NS2F?

It is a 1U, 64-port 400Gb/s InfiniBand switch in the NVIDIA Quantum-2 QM9700 family, with 32 OSFP connectors, 51.2 Tb/s aggregate throughput, external UFM management and power-to-connector (P2C) forward airflow.

How many ports and what throughput does it provide?

64 × 400Gb/s non-blocking InfiniBand ports, with aggregate bidirectional throughput up to 51.2 Tb/s and more than 66.5 billion packets per second capacity. The 32 OSFP connectors can support up to 128 ports of 200Gb/s through port-split technology.

Is it internally or externally managed?

The MQM9790-NS2F is the externally managed model in the QM9700 family and can utilize NVIDIA UFM feature sets. The internally managed QM9700 model features an on-board subnet manager for out-of-the-box bring-up of up to 2,000 nodes; that on-board subnet manager applies to the QM9700, not the QM9790.

What is the difference between NS2F and NS2R?

The main difference is airflow direction: NS2F uses P2C (forward) airflow, while NS2R uses C2P (reverse) airflow. Confirm against your rack design before ordering.

Does it support NVIDIA SHARPv3?

The QM9700 400Gb/s InfiniBand switches extend NVIDIA In-Network Computing and introduce SHARPv3, which allows 32X higher AI acceleration power compared to the previous generation.

What power and cooling does it require?

It uses 1+1 redundant hot-swappable power supplies with a 200–240Vac input range, 80 Gold+ and ENERGY STAR certified, plus a hot-swappable fan unit. Operating temperature is 0°C–40°C.

Can you confirm compatibility with my optical modules and cables?

Yes. Please share your planned OSFP optical modules, copper or fiber cables and host adapter models, and our technical team will confirm compatibility before ordering.