NVIDIA Quantum-2 QM9700 Series Switches
Scaling Out Data Centers with NDR 400G InfiniBand Smart Switches
In a 1U conventional chassis configuration, the NVIDIA Quantum 2-based QM9700 and QM9790 switch systems offer an unparalleled 64 ports of NDR 400Gb/s InfiniBand per port. With an unprecedented capacity of over 66.5 billion packets per second (BPPS), a single switch can handle an aggregated bidirectional throughput of 51.2 terabits per second (Tb/s). NVIDIA Quantum-2, which supports the most recent NDR technology, is a high-speed, incredibly low-latency, and scalable solution that integrates cutting-edge technologies like adaptive routing, NVIDIA Scalable Hierarchical Aggregation and Reduction Protocol (SHARP), and Remote Direct Memory Access (RDMA).
In contrast to other networking solutions, NVIDIA InfiniBand offers the highest total application throughput with improved virtual lane (VL) mapping, congestion control, quality of service (QoS), and self-healing network capabilities. The NVIDIA Quantum-2 QM9700 and QM9790 NDR InfiniBand fixed-configuration switches are the perfect rack-mounted InfiniBand solution because they offer a range of topologies, such as Fat Tree, SlimFly, DragonFly+, multi-dimensional Torus, and more, giving you the most flexibility possible. They also support a wide range of software ecosystems and are backwards compatible with earlier generations.
The Era of Data-Driven Computing
High-resolution simulations, enormous datasets, and intricate, highly parallelized algorithms that must communicate in real time must all be processed incredibly quickly for sophisticated research. The third iteration of NVIDIA SHARP technology, known as SHARPv3, is introduced by the NVIDIA Quantum-2 QM9700 NDR InfiniBand switches, which also expand NVIDIA In-Network Computing capabilities. In comparison to the previous generation, SHARPv3 offers 32X more AI acceleration capability and supports up to 64 parallel flows, providing nearly infinite scalability for huge data aggregation across the network.
World-Class InfiniBand Performance
Network Computing
By shifting collective communication tasks from the CPU to the switch network, NVIDIA Mellanox Scalable Hierarchical Aggregation and Reduction Protocol (SHARP) In-Network Computing technology increases application performance by an order of magnitude.
Self-Healing Computing
The self-healing networking features of the NVIDIA Mellanox InfiniBand improve system performance, scalability, and network usage by overcoming link failures and recovering the network 5,000X quicker than any software-based solution.
UFM Management
NVIDIA Mellanox Unified Fabric Management (UFM) solutions reduce OPEX while achieving greater utilization of fabric resources and a competitive edge by combining improved, real-time network telemetry with AI-powered cyber intelligence and analytics.
Key Features NVIDIA Quantum-2 QM9700
InfiniBand Performance Leads the Industry
By offering up to 400 Gbps InfiniBand connectivity per port, the NVIDIA Quantum-2 QM9700 Series doubles the bandwidth of earlier models. For applications requiring a lot of data, this guarantees extremely low latency and great throughput.
Improved Scalability
Massive scaling of AI models and HPC applications is made possible by the NVIDIA Quantum-2 QM9700 Series’ sophisticated InfiniBand networking capabilities. It reduces bottlenecks in distributed computing systems by efficiently supporting multi-node clustering.
Complex In-Network Processing
Performance in AI training and HPC workloads is greatly enhanced by the NVIDIA Quantum-2 InfiniBand’s support for SHARPv3 (Scalable Hierarchical Aggregation and Reduction Protocol), which transfers collective communication tasks from the CPU/GPU to the network fabric.
Features of Integrated Security
End-to-end encryption is a feature of the QM9700 Series that enables safe communication between nodes in HPC and AI clusters. This preserves excellent performance while protecting sensitive data.
Architecture with Energy Efficiency
NVIDIA has reduced the energy usage per bit sent by optimizing the Quantum-2 series’ power efficiency. By doing this, data centers are able to maintain optimal performance while reducing operating expenses.
Applications of NVIDIA Quantum-2 QM9700
Deep Learning and AI Training
The NVIDIA Quantum-2 QM9700‘ high bandwidth and low latency networking make it perfect for training large-scale AI models, such as neural networks, generative AI, and large language models (LLMs).
Scientific Computing and Simulations
The Quantum-2 InfiniBand speeds up intricate simulations in physics, engineering, genetics, and climate modeling, giving researchers quicker insights and more precise findings.
Cloud Computing and Enterprise Data Centers
Businesses are increasingly embracing cloud-based HPC, real-time processing, and AI-driven analytics. The QM9700 Series offers the infrastructure required for high-performance cloud networking.
Competitive Advantage
For AI and HPC workloads, the NVIDIA Quantum-2 QM9700 provides superior efficiency, higher throughput, and lower latency than conventional Ethernet-based solutions. It has an advantage in performance-critical applications because of its SHARPv3 offloading and in-network computing capabilities, which enable faster data flow and computation.
