NVIDIA MQM9700-NS2F 400Gb/s InfiniBand Smart Switch 64-Port Data Center Networking

The NVIDIA Quantum-2 MQM9700-NS2F is a cutting-edge, high-radix InfiniBand network switch engineered for next-generation data centers. Delivering 64 non-blocking ports of 400Gb/s NDR InfiniBand throughput in a standard 1U form factor, this network card equivalent in switching power is designed to meet the escalating demands of artificial intelligence (AI), high-performance computing (HPC), and massive-scale data analytics. It aggregates a remarkable 51.2 Tb/s of bidirectional bandwidth, establishing a new benchmark for data center fabric performance and efficiency. The mq9700 series, including this model, represents a pivotal advancement in network hardware, integrating smart switching and in-network computing to accelerate workloads and simplify infrastructure.

• Unmatched Density & Performance: 64 ports of 400Gb/s InfiniBand (32 OSFP connectors) with an aggregate throughput of 51.2 Tb/s.
• In-Network Computing: Features 3rd-gen NVIDIA SHARPv3 technology, accelerating collective operations and reducing data movement by up to 32x compared to prior generations.
• Intelligent Fabric: Built-in subnet manager for out-of-the-box orchestration of up to 2,000 nodes.
• Topology Flexibility: Supports Fat Tree, DragonFly+, SlimFly, and multi-dimensional Torus topologies, enabling scalable and efficient network designs.
• High Reliability: 1+1 redundant, hot-swappable power supplies and fan units with forward (P2C) airflow design.
• Advanced Software: Managed by NVIDIA MLNX-OS, offering comprehensive control via CLI, WebUI, SNMP, and JSON interfaces.

The mq9700 platform leverages several industry-defining technologies to deliver a superior network switch solution:

• NDR 400Gb/s InfiniBand: Provides ultra-low latency and extreme bandwidth for data-intensive workloads.
• SHARPv3 (Scalable Hierarchical Aggregation and Reduction Protocol): Performs mathematical operations within the network hardware, drastically boosting AI and HPC application performance.
• Adaptive Routing & Congestion Control: Dynamically optimizes data paths to maximize throughput and minimize latency.
• Remote Direct Memory Access (RDMA): Enables direct memory-to-memory data transfer, bypassing the CPU to reduce overhead.
• Port-Split Technology: Allows flexible configuration of 128 ports at 200Gb/s, offering cost-effective topology options.

The MQM9700-NS2F network switch operates as the intelligent core of a data center fabric. Data packets enter through the high-speed 400Gb/s OSFP ports. The integrated switch ASIC, guided by MLNX-OS and the onboard subnet manager, performs advanced packet forwarding, adaptive routing, and congestion management. Crucially, with SHARPv3 enabled, the switch can offload collective communication operations (like reductions and aggregations) from the compute servers, processing them "in-network" as data flows through. This principle significantly reduces the volume of data traversing the network and the computational load on end-point servers, leading to faster job completion times and higher overall system efficiency for parallel applications.

This high-performance nic card alternative in switching form is ideal for environments requiring extreme-scale data movement and computational efficiency:

• AI & Machine Learning Training Clusters: Connects thousands of GPUs, enabling fast model training with efficient all-reduce operations via SHARPv3.
• High-Performance Computing (HPC): Powers scientific research involving large-scale simulations (climate, genomics, astrophysics).
• Hyperscale Data Analytics: Accelerates data processing pipelines for real-time analytics on massive datasets.
• Cloud and Enterprise Data Centers: Serves as a high-density top-of-rack (ToR) or spine switch for building scalable, low-latency fabrics.

• Performance Leadership: Offers the highest port density and throughput (51.2 Tb/s) in a 1U network switch, outperforming competing Ethernet and InfiniBand solutions.
• In-Network Computing Pioneer: Exclusive SHARPv3 technology transforms the network card from a passive data pipe into an active compute resource, delivering unparalleled application acceleration.
• Design Flexibility: Port-split capability (to 128x200Gb/s) and support for multiple topologies reduce total cost of ownership (TCO) and allow tailored deployments.
• Proven Reliability: Features like self-healing networks, adaptive routing, and QoS ensure robust operation in demanding, large-scale environments.
• Comprehensive Ecosystem: Backward compatibility and broad software support ensure easy integration into existing and new data centers.

• Warranty: Standard 1-year limited hardware warranty.
• Technical Support: Access to NVIDIA's global technical support and resources.
• Software Updates: Ongoing firmware and MLNX-OS software updates provided.
• Certification & Compliance: Product is certified for safety (cTUVus, CE, CU), EMC (FCC, CE), and environmental (RoHS) standards.

A: It is primarily designed as the foundational network hardware for large-scale AI training clusters and HPC systems where extreme bandwidth, ultra-low latency, and in-network computing acceleration are critical.

A: Unlike a traditional network switch that only forwards packets, SHARPv3 allows the mq9700 to perform mathematical operations (like summation for AI gradients) on data as it passes through, drastically reducing network traffic and accelerating collective tasks.

A: Yes. The switch is backward compatible with previous generations of InfiniBand. Using appropriate cables or adapters, you can connect to 200Gb/s (HDR) or 100Gb/s (EDR) endpoints, making it a future-proof investment.

A: The managed MQM9700-NS2F has a built-in subnet manager for fabric configuration and can operate standalone. The unmanaged MQM9790 version relies on an external subnet manager, typically the NVIDIA UFM software platform, for centralized management of larger fabrics.

• Compatibility: Ensure servers and storage systems are equipped with compatible 400/200/100Gb/s InfiniBand host channel adapters (nic cards).
• Environmental Conditions: Operate within specified temperature (0°C to 40°C) and humidity (10% to 85% non-condensing) ranges. Ensure adequate airflow (P2C for NS2F model) in the rack.
• Power: Connect to a stable 200-240V AC source. Utilize both power supplies for redundancy.
• EMC & Safety: Installation must comply with local regulations for electromagnetic compatibility (EMC) and electrical safety. The product is certified for major global markets.
• Maintenance: Only qualified personnel should perform installation, maintenance, or component replacement (fans, power supplies).

With over a decade of industry expertise, we maintain extensive manufacturing capabilities and are supported by a robust technical engineering team. We have established successful partnerships with numerous global clients, delivering competitive pricing structures and superior service quality. Our comprehensive product portfolio includes leading technology brands such as Mellanox, Ruckus, Aruba, and Extreme Networks. We maintain substantial inventory of network infrastructure components including high-performance switches, network interface controllers, wireless access points, network management controllers, and structured cabling solutions, ensuring reliable supply chain management and timely delivery performance.