MQM9790-NS2F 64-Port 400Gb/s InfiniBand Smart Switch

Unmanaged NDR Switch | 51.2 Tb/s Aggregate Throughput | For External Subnet Management

Engineered for extreme-scale AI, HPC, and hyperscale data centers. The MQM9790-NS2F delivers 64 non-blocking ports of 400Gb/s InfiniBand in a compact 1U chassis, enabling flexible topologies and massive scalability with SHARPv3 in-network computing.

Hong Kong Starsurge Group presents the NVIDIA Quantum-2 MQM9790-NS2F — a high-performance, unmanaged 400Gb/s InfiniBand switch designed for customers who require external subnet management (e.g., NVIDIA UFM). As part of the QM9700 series, this switch provides 32 OSFP ports supporting 64* 400Gb/s connections (or up to 128 ports at 200Gb/s via split technology). With a landmark 51.2 terabits per second bidirectional throughput and over 66.5 billion packets per second capacity, the MQM9790-NS2F is ideal for research institutions, finance, manufacturing, and enterprise AI clusters demanding low latency and lossless fabric.

• Unprecedented Port Density: 64 ports of 400Gb/s NDR InfiniBand in 1U, non-blocking architecture.
• Double-Density 200Gb/s Mode: Supports up to 128 ports of 200Gb/s using NVIDIA port-split technology, reducing TCO for leaf-spine topologies.
• In-Network Computing: SHARPv3 (Scalable Hierarchical Aggregation and Reduction Protocol) delivers 32x higher AI acceleration compared to prior generation.
• Advanced Fabric Services: RDMA, adaptive routing, congestion control, enhanced VL mapping, and self-healing network capabilities.
• Unmanaged Model: Designed for external Subnet Manager (UFM or on-board SM on another device) — maximum flexibility for large fabrics.
• Cooling & Redundancy: Front-to-rear (P2C) airflow, 1+1 redundant hot-swappable power supplies, and hot-swappable fans.
• Backward Compatible: Supports previous InfiniBand generations (HDR, EDR, FDR).

Built on the NVIDIA Quantum-2 ASIC, the MQM9790-NS2F leverages state-of-the-art 400Gb/s SerDes technology. It incorporates Remote Direct Memory Access (RDMA) for low CPU overhead and high throughput, adaptive routing to avoid fabric hotspots, and NVIDIA SHARPv3 for in-network reductions, dramatically accelerating collective operations in MPI and AI frameworks. The switch also includes advanced congestion control and enhanced QoS to guarantee performance for mission-critical workloads.

With support for multiple topologies — including Fat Tree, SlimFly, DragonFly+, and multi-dimensional Torus — the MQM9790-NS2F enables architects to build cost-effective, highly resilient networks for next-gen supercomputing.

• AI & Machine Learning Clusters: High-speed GPU interconnects for large language model training and inference.
• High-Performance Computing (HPC): Research simulations, weather modeling, genomics, and quantum computing labs.
• Hyperscale Data Centers: Spine-leaf architectures with massive east-west traffic.
• Enterprise Cloud & Finance: Low-latency trading systems and real-time analytics.
• Government & Education: National labs, university supercomputing centers requiring deterministic performance.

Fully interoperable with NVIDIA InfiniBand ecosystem: ConnectX-6/7 adapters, LinkX cables, and Quantum-2 series switches. Backwards compatible with HDR, EDR, and FDR devices. Supports major Linux distributions (RHEL, Ubuntu, Rocky Linux) and Windows Server with appropriate drivers. The unmanaged operation requires an external Subnet Manager (e.g., NVIDIA UFM or OpenSM running on a server).

For P2C airflow (front-to-rear), cold aisle is at the power supply side. Confirm your data center cooling scheme before ordering.

• Cost-efficient unmanaged operation: Ideal for larger deployments where a centralised external Subnet Manager (UFM) reduces per-switch overhead.
• Highest radix in 1U: 64x 400G ports minimizes switch tiers, lowers latency, and simplifies cabling.
• Energy efficient: Gold+ power supplies and smart fan control reduce OPEX.
• Flexible cabling: OSFP connectors support passive/active copper and optical cables, including active fiber.
• Future-ready scalability: Support for SHARPv3 and adaptive routing ensures investment protection for next-gen AI frameworks.

Hong Kong Starsurge Group provides end-to-end support including pre-sales consulting, integration services, and global logistics. Our technical team offers deployment assistance, RMA services, and extended warranty options. For volume orders, we deliver customized cabling and configuration validation. Multilingual support available for APAC, EMEA, and Americas regions.