NVIDIA Quantum-2 QM9700

NVIDIA Quantum-2 QM9700: High-Performance Networking for AI and HPC

The NVIDIA® Quantum-2 QM9700 InfiniBand switch is an advanced high-performance networking platform designed to meet the most demanding requirements of modern artificial intelligence (AI), high-performance computing (HPC), and hyperscale cloud data centers.

Equipped with 64 ports of 400Gb/s OSFP interfaces and built on NDR (Next Data Rate) technology, the QM9700 delivers unmatched scalability, low latency, and extreme bandwidth, setting a new standard for next-generation data center fabrics.

Key Features at a Glance

• Data Rate: NDR (400Gb/s per port)
• Ports: 64 × 400Gb/s
• Network Interface: OSFP (Octal Small Form-factor Pluggable)
• Technology: InfiniBand
• Performance Class: Extreme throughput designed for Exascale AI workloads

Designed for Exascale AI and Scientific Research

The QM9700 is part of the NVIDIA Quantum-2 InfiniBand family, a platform tailored for environments where data transfer speed and latency are critical.

It enables AI developers and researchers to solve complex problems in record time by maximizing data movement efficiency and reducing communication bottlenecks.

Whether powering large AI models in training clusters or accelerating simulations in scientific computing, the QM9700 ensures optimal utilization of computational resources through ultra-high bandwidth and sophisticated offload capabilities.

In-Network Computing: Accelerating Performance Beyond Traditional Limits

A defining innovation of the Quantum-2 series is NVIDIA In-Network Computing, which pushes computation into the fabric itself to reduce CPU and GPU overhead.

The QM9700 leverages pre-configured and programmable acceleration engines, enabling a new level of performance through features such as:

• NVIDIA SHARP™ (Scalable Hierarchical Aggregation and Reduction Protocol): Hardware-based data aggregation that minimizes latency and CPU usage in operations like reduce, all-reduce, and broadcast.
• MPI Tag Matching and Collectives: Improved efficiency in message passing applications, including hardware acceleration for collective operations like MPI_Alltoall.
• Programmable Packet Processing Cores: Customize data handling within the switch for enhanced flexibility in complex workloads.
• Full Transport Offload:
  • RDMA (Remote Direct Memory Access): Reduces CPU overhead by enabling direct memory access between systems.
  • GPUDirect RDMA and GPUDirect Storage: Ensures seamless data movement between GPUs and storage subsystems without involving the host CPU.

These enhancements lead to better ROI (return on investment) and lower total cost per node, making QM9700 ideal for hyperscale deployments.

Scalable Architecture for Modern Data Centers

With 64 OSFP ports, each supporting 400Gb/s speeds, the QM9700 offers up to 25.6 Tb/s of switching capacity.

Its modular and scalable design enables seamless expansion of AI clusters and HPC environments.

Key architectural benefits include:

• Linear Scalability: Add more nodes without loss of performance or increased latency.
• Ultra-Low Latency Switching: Less than 90 nanoseconds port-to-port, enabling real-time responsiveness in high-throughput environments.
• End-to-End Congestion Control: Intelligent fabric management to prevent packet drops and maintain high application performance.

Advanced Networking Interface: OSFP

The use of OSFP (Octal Small Form-factor Pluggable) modules offers superior thermal management and bandwidth capacity.

These interfaces are ideal for ultra-high-speed data center applications where power density and signal integrity are critical.

Advantages of OSFP:

• Enhanced cooling for high-speed optical modules
• Future-proof design supporting next-gen transceivers
• Interoperability across NVIDIA’s InfiniBand ecosystem

Product Variants and Configurations

The QM9700 switch comes in multiple product SKUs, tailored to suit different deployment needs while maintaining core performance capabilities.

Below is an overview of available configurations:

SKU	OPN	Data Rate	Network Interface	Ports	Technology
920-9B210-00FN-0M0	MQM9700-NS2F	NDR	OSFP	32	InfiniBand
920-9B210-00RN-0M2	MQM9700-NS2R	NDR	OSFP	32	InfiniBand
920-9B210-00FN-0D0	MQM9790-NS2F	NDR	OSFP	32	InfiniBand
920-9B210-00RN-0D0	MQM9790-NS2R	NDR	OSFP	32	InfiniBand

Each variant is designed to integrate seamlessly with Quantum-2 systems and offers the same set of software-defined features and In-Network Computing capabilities.

Exceptional ROI and Cost-Efficiency

One of the primary challenges in building AI and HPC clusters is balancing performance with infrastructure costs.

The NVIDIA Quantum-2 QM9700 mitigates this concern by delivering:

• Superior performance-per-dollar: Reduce the number of switches needed in a fabric while increasing throughput.
• Energy Efficiency: Low power consumption per Gb/s reduces overall TCO (Total Cost of Ownership).
• Software-Defined Flexibility: Easily adapt to evolving workloads without hardware upgrades.

These advantages make the QM9700 a long-term investment for organizations aiming for sustained innovation and competitive advantage.

Ideal Applications and Use Cases

The QM9700 is engineered for deployment in:

• Exascale AI Clusters: Supports massive model training and inference at scale.
• HPC Environments: Accelerates computation-heavy scientific workloads in physics, genomics, weather modeling, etc.
• Hyperscale Cloud Platforms: Serves as a backbone for cloud-native services with high network throughput.
• Real-Time Data Analytics: Ensures consistent performance in latency-sensitive applications.

Conclusion: Ready for the Future of High-Speed Networking

The NVIDIA Quantum-2 QM9700 is more than just a network switch it’s an intelligent data acceleration platform that redefines the performance boundaries of AI and HPC infrastructures.

With NDR speeds, 64 ultra-fast OSFP ports, and cutting-edge In-Network Computing technologies, the QM9700 is purpose-built to meet the explosive demands of tomorrow’s data-driven world.

Whether you’re building the next exascale AI system or optimizing performance in a global cloud network, the QM9700 delivers the scalability, performance, and innovation you need today and into the future.