The NVIDIA Quantum-2 QM9700 (SKU 920-9B210-00FN-0M0) is a high-performance 1U InfiniBand switch designed for enterprise data center and HPC deployments requiring ultra-high bandwidth and low-latency interconnect. It delivers 64 ports of NDR 400Gb/s per port with a non-blocking architecture achieving 51.2Tbps aggregate throughput. The switch features 32 OSFP connector cages supporting passive copper or active fiber cabling, dual 1+1 redundant hot-swappable power supplies (200-240VAC input), and managed power-to-connector (P2C) forward airflow cooling. Advanced networking capabilities include Quality of Service (QoS), RDMA support, adaptive routing, congestion control, self-healing network features, and NVIDIA SHARP (Scalable Hierarchical Aggregation and Reduction Protocol) for optimized collective communications. Management interfaces include dual RJ-45 Ethernet ports, USB 3.0, and USB I2C connectivity. The system runs MLNX-OS with X86 Coffee Lake i3 CPU, 8GB DDR4 SO-DIMM memory, and 16GB M.2 SATA SSD storage. Operating temperature range is 0-40°C (forward airflow) with non-operational range -40 to 70°C. Fully compliant with RoHS, CB, cTUVus, CE, and CU safety certifications.
64 NDR 400Gb/s ports with 51.2Tbps aggregate throughput.
32 OSFP connectors supporting passive copper and active fiber.
Dual 1+1 redundant hot-swappable power supplies (200-240VAC).
RDMA, QoS, adaptive routing, and NVIDIA SHARP protocol support.
1U rack-mountable form factor with managed P2C forward airflow.
X86 Coffee Lake i3 CPU with 8GB DDR4 and 16GB M.2 SSD storage.
Expedited and overnight shipping options are available at checkout, priced by destination and service.
Purchase orders accepted from qualified institutions.
Refunds within 14 days and replacements within 30 days of delivery with an RMA (25% restocking fee on customer-initiated returns), and every part carries our Standard 30-Day Warranty - see our return policy or contact US IT Hardware.
Featuring
Manufacturer part #
920-9B210-00FN-0M0
Brand
NVIDIA
Condition
New
Model Name
QM9700 1U NDR 400Gbps InfiniBand Switch
Port Speed
400Gb/s per port (NDR), supports 40/56/100/200/400Gb/s
Energy Star
Energy Star certified power supplies
NVIDIA SHARP
Scalable Hierarchical Aggregation and Reduction Protocol for hardware-accelerated collectives
Product Line
NVIDIA Quantum-2 InfiniBand
Product Type
High-Performance InfiniBand Switch
RDMA Support
Full RDMA over InfiniBand (RoIB) support
System Memory
Compatible systems
The QM9700 is fully compatible with NVIDIA DGX SuperPOD infrastructure, standard InfiniBand ecosystems, and any NDR 400Gb/s InfiniBand-capable compute nodes. Requires standard data center power (200-240VAC), cooling, and 19-inch rack infrastructure. Management via MLNX-OS or standard network protocols.
NVIDIA DGX SuperPOD
Model: DGX B300
Form factor: Data Center
Generation: Current
Validated as NDR switch for DGX SuperPOD with DGX B300 systems in reference architecture deployments.
NVIDIA Quantum-2 InfiniBand
Model: QM9700, QM9790
Form factor: 1U Switch
Generation: Quantum-2
Part of NVIDIA Quantum-2 InfiniBand switch family; compatible with QM9790 for multi-switch fabric deployments.
Generic InfiniBand HCA
Model: Any NDR 400Gb/s InfiniBand Host Channel Adapter
Form factor: PCIe
Generation: NDR
Compatible with any standards-compliant NDR 400Gb/s InfiniBand HCA supporting OSFP connectors.
Best for
High-Performance Computing (HPC) Clusters
Interconnect for large-scale scientific computing, simulation, and modeling environments requiring ultra-low latency and massive bandwidth for inter-node communication across hundreds of compute nodes.
Details
Shipping, warranty & returns
Customer reviews
No reviews yet — be the first to review this part.
Free ground shipping over $100
Secure SSL checkout. Visa, Mastercard, Amex and Discover accepted. ACH / wire available for businesses.
Buy with confidence
New, certified-refreshed and OEM stock with guaranteed compatibility, backed by warranty on every part.
Passive copper (up to 10m), active copper, active fiber optical
Management Interfaces
2x RJ-45 Ethernet, 1x USB 3.0, 1x USB I2C
Data Center Spine/Core Switching
Enterprise data center core or spine layer switch aggregating multiple top-of-rack (ToR) switches, supporting east-west traffic patterns and enabling non-blocking fabric architectures.
AI/ML Training Infrastructure
GPU cluster interconnect for distributed deep learning and large language model training, where RDMA and SHARP protocol acceleration reduce collective communication overhead.
Storage Area Networks (SAN)
High-bandwidth storage fabric for NVMe-oF and iSER deployments requiring consistent low-latency access to shared storage resources across multiple servers.
Financial Services & Real-Time Analytics
Ultra-low latency trading and analytics platforms requiring deterministic packet delivery and minimal jitter for time-sensitive applications.
Supercomputing Centers
Backbone interconnect for national laboratories and research institutions deploying petascale or exascale computing systems with thousands of compute nodes.
Frequently asked questions
What is the total switching capacity and port configuration of the QM9700?
The QM9700 delivers 64 ports of NDR 400Gb/s per port with a non-blocking architecture, providing 51.2Tbps aggregate switching capacity. Each port supports 32 OSFP connectors that accept passive copper or active fiber cabling, enabling flexible deployment in various data center topologies.
How does NVIDIA SHARP protocol improve collective communications in HPC clusters?
SHARP (Scalable Hierarchical Aggregation and Reduction Protocol) offloads collective operations (AllReduce, AllGather, Broadcast) from CPU to the switch hardware, reducing latency and CPU overhead. This is critical for distributed AI training and scientific computing where collective communications represent 20-40% of application runtime.
What are the power consumption specifications and cooling requirements?
Maximum power consumption is 1,084W with passive cables or 1,720W with optical cables consuming 18.5W per port. The switch uses managed power-to-connector (P2C) forward airflow cooling with dual 1+1 redundant hot-swappable power supplies rated for 200-240VAC input, 80+ Gold certified, and Energy Star compliant.
Is the QM9700 compatible with existing InfiniBand infrastructure?
Yes, the QM9700 is fully compatible with InfiniBand ecosystem standards and supports RDMA over InfiniBand (IB). It integrates with existing IB-based HPC clusters, storage networks, and GPU clusters. Passive copper cables up to 10m and active fiber cables support mixed-generation deployments.
What management and monitoring capabilities are included?
The QM9700 includes dual RJ-45 Ethernet management ports, USB 3.0, and USB I2C connectivity. It runs MLNX-OS with embedded X86 Coffee Lake i3 CPU, 8GB DDR4 memory, and 16GB M.2 SSD for local storage. Management features include telemetry, health monitoring, and firmware updates via standard network interfaces.
What environmental operating conditions and certifications apply?
Operating temperature range is 0-40°C with forward airflow or 0-35°C with reverse airflow; non-operational range is -40 to 70°C. Humidity tolerance is 10-85% non-condensing (operational) and 10-90% non-condensing (non-operational). Altitude rating up to 3,050m. Certified RoHS, CB, cTUVus, CE, CU, and S-Mark compliant.
How does the QM9700 compare to previous-generation InfiniBand switches?
The QM9700 is based on NVIDIA Quantum-2 architecture delivering 400Gb/s per port versus 200Gb/s on Quantum-1 (HDR). It provides 2x bandwidth density, improved power efficiency, enhanced SHARP capabilities, and better congestion control. The 1U form factor maintains compatibility with standard data center rack infrastructure.
Cisco MerakiMS125-48LP-HW
Allied TelesisAT-8624T/2M
Allied TelesisAT-9000/24
Dell469-3414
Allied TelesisAT-GS910/24-50
Allied TelesisAT-FS717FC/SC-10
Allied TelesisAT-9424T/SP