NVIDIA Quantum-2 MQM9790-NS2F 400Gb/s InfiniBand Smart Switch

Unprecedented 64 ports of 400Gb/s NDR InfiniBand in a 1U form factor—delivering 51.2Tb/s aggregate throughput and 66.5 billion packets per second with SHARPv3 in-network acceleration for extreme-scale AI and HPC environments.

• 64 non-blocking 400Gb/s InfiniBand ports (32 OSFP connectors with double-density radix)
• Aggregate switching capacity: 51.2 Tb/s (bidirectional) | Packet processing: 66.5 BPPS
• Third-generation SHARP technology (SHARPv3) for 32x higher AI acceleration
• Unmanaged SKU (MQM9790-NS2F) with on-board subnet manager supporting up to 2000 nodes
• P2C airflow, 1+1 redundant power supplies, 1U standard depth, backward compatible with HDR/EDR

Characteristics

• 400Gb/s NDR Performance: 64 ports at 400Gb/s per port with full bidirectional bandwidth and ultra-low latency.
• Double-Density 200G Mode: Port-split technology yields 128 ports of 200Gb/s—ideal for dense ToR deployments.
• SHARPv3 In-Network Computing: Hardware-based data aggregation/reduction during collective communications.
• Advanced Topology Support: Fat Tree, SlimFly, DragonFly+, multi-dimensional Torus, and adaptive routing.
• Energy & Space Optimized: 80 PLUS Gold certified, hot-swappable fans, 1U chassis with front-to-rear or rear-to-front cooling options.

Technology & Standards

Based on NVIDIA Quantum-2 InfiniBand architecture, the MQM9790-NS2F integrates SHARPv3 for in-network reductions, RDMA (Remote Direct Memory Access), adaptive routing, enhanced virtual lane (VL) mapping, and congestion control. It supports NDR 400Gb/s and is backward compatible with HDR100, HDR, EDR, and FDR. The switch runs MLNX-OS (unmanaged version relies on external UFM or OpenSM for advanced fabric management).

Working Principle: In-Network Computing with SHARPv3

Unlike conventional switches that only forward packets, the MQM9790-NS2F performs computations on data as it traverses the fabric. Using dedicated silicon, it aggregates and reduces data for collective operations (e.g., all-reduce, barrier) at wire speed, drastically reducing traffic between endpoints. This enables linear scalability for AI training clusters and HPC simulations. The embedded x86 Coffee Lake i3 processor manages fabric initialization and topology discovery, while data acceleration occurs in dedicated hardware.

Applications & Deployment

• Exascale AI Training: Large language models and deep learning requiring massive all-reduce bandwidth.
• High-Performance Computing (HPC): Climate modeling, molecular dynamics, and computational fluid dynamics.
• Double-Density Top-of-Rack (ToR): 128-port 200G configuration for dense server connectivity.
• Cloud & Enterprise Data Centers: High-throughput storage (NVMe-oF), database acceleration.
• Research & Government Labs: Ultra-low-latency fabric for scientific discovery.

Technical Specifications & Ordering Options

Note: All models include x86 Coffee Lake i3 CPU, 8GB DDR4, M.2 SSD 16GB, 1x USB 3.0, 1x I²C USB, 1x RJ45 (UART). Weight: ~14.5kg. Cables: OSFP passive/copper/active optical.

Advantages & Differentiators

• vs. Previous Gen (QM8700): 2.5x more ports (64 vs. 40), 3.2x aggregate bandwidth (51.2Tb/s vs. 16Tb/s), and SHARPv3 delivering 32x higher AI acceleration.
• vs. Competitive 400G Switches: Integrated SHARPv3 reduces job completion times; double-density 200G mode lowers network tiers and saves cost.
• Densest ToR Solution: 128 ports of 200G in 1U—ideal for high-density racks.
• Energy Proportional: Power scales with port utilization; ENERGY STAR certified.
• Seamless Backward Compatibility: Works with ConnectX-6/5/4 network cards and existing cabling.

Service & Support

We offer full lifecycle support: 24/7 technical consultation, RMA handling, and on-site assistance. Our team provides topology design, firmware compatibility validation, and integration with existing InfiniBand fabrics. All units carry a 1-year warranty (extendable). Orders ship within 24-48 hours from our 10M+ inventory. Custom cabling and network card bundles available.

Frequently Asked Questions (FAQ)

Q: Is the MQM9790-NS2F compatible with Mellanox ConnectX-6 network cards?

A: Yes, it fully supports ConnectX-6 (HDR/EDR) and ConnectX-5/4. For 400Gb/s operation, use ConnectX-6 or newer network cards that support NDR.

Q: How does the 128-port 200Gb/s mode work?

A: Each OSFP connector houses 8 lanes. By splitting a 400G (8 lanes) port into two independent 200G (4 lanes) connections, the switch effectively doubles the port count for 200G endpoints—ideal for high-density ToR.

Q: Can this unmanaged switch run a subnet manager?

A: Yes, it includes an onboard subnet manager capable of managing up to 2000 nodes out-of-the-box. For advanced features like UFM, connect an external server running OpenSM or NVIDIA UFM.

Q: What cooling options are available?

A: The MQM9790-NS2F has P2C (port-to-power/forward) airflow. If your rack uses rear-to-front cooling, order MQM9790-NS2R (C2P). Hot-swappable fan units support both directions.

Q: What is the typical power consumption at 50% load?

A: Depending on utilization and cable types, typical power draw ranges from 250W to 350W at half load. The switch dynamically reduces power when ports are idle.

Precautions & Compatibility Notes

• Airflow Direction: Verify rack cooling matches your model: P2C (MQM9790-NS2F) or C2P (MQM9790-NS2R). Improper airflow may cause overheating.
• Power Supply: Input range 200-240V AC only (not 100-127V). Use both power feeds for redundancy.
• Cabling: For 400G, use OSFP passive copper (up to 3m) or active optical modules. 200G splitting requires OSFP to 2x200G breakout cables.
• Operating Environment: Temperature 0°C–40°C, humidity 10%–85% non-condensing, altitude ≤3050m. EMC compliant with CE, FCC, VCCI, ICES, RCM.
• Software: External subnet manager/UFM recommended for large-scale fabrics. Always update firmware in coordination with network card versions.

Company Introduction

With over a decade of experience, we operate a large-scale factory backed by a strong technical team. Our extensive customer base and domain expertise enable us to offer competitive pricing without compromising on quality. As authorized distributors for Mellanox, Ruckus, Aruba, and Extreme, we stock original network switches, network card (nic card) solutions, wireless Access Points, controllers, and cabling. We maintain a 10 million USD inventory to ensure rapid fulfillment across diverse product lines. Every shipment is verified for accuracy, and we provide 24/7 consultation and technical support. Our professional sales and technical teams have earned a high reputation in global markets—partner with us for reliable infrastructure solutions.