Limited Time Offer: Save $350 on your Ai Cluster → Shop Now

--

Days

--

Hours

--

Mins

--

Secs

Sale Ended

NVIDIA Quantum-2 QM9700 NDR InfiniBand Switch, 64× 400G
NVIDIA Quantum-2 QM9700 NDR InfiniBand Switch, 64× 400G

Vendor:NVIDIA

NVIDIA Quantum-2 QM9700 NDR InfiniBand Switch, 64× 400G

$44,462.00

The NVIDIA Quantum-2 QM9700 is a 1U NDR InfiniBand switch built for large-scale AI, HPC, research, and data center fabrics. It provides 64 non-blocking 400Gb/s InfiniBand ports through 32 OSFP connectors, delivering up to 51.2Tb/s of aggregate bidirectional throughput in a compact rackmount design.

In stock, ready to ship
Product datasheet

Pairs well with

30-Day AISuperStore Guarantee 

TECHNOLOGY AT A GLANCE

Explore the capabilities

51.2Tb/s of switching capacity

64 non-blocking 400Gb/s InfiniBand ports deliver massive aggregate bandwidth in a single 1U chassis.

In-network AI acceleration

NVIDIA SHARPv3 performs collective operations within the network to reduce data movement and accelerate distributed workloads.

400G or 200G flexibility

Use 32 OSFP connectors for 64× 400G links or port-split technology for up to 128× 200G connections.

Built for continuous operation

Redundant hot-swappable power supplies and 6+1 hot-swappable fans simplify maintenance in dense rack deployments.

The fabric behind extreme-scale AI

The Quantum-2 QM9700 brings 400G NDR InfiniBand connectivity, in-network computing, and dense 1U deployment to the center of your AI and HPC infrastructure. Build high-bandwidth fabrics that keep GPUs, storage, and compute nodes exchanging data at scale.

High-density 400G fabric connectivity

The QM9700 concentrates an extreme amount of InfiniBand bandwidth into a compact 1U switch, with flexible high-speed connectivity and serviceable infrastructure components.

  • 32 OSFP connectors: Each physical connector supports two 400Gb/s NDR InfiniBand links, providing 64 non-blocking 400G ports in total.
  • 200G port splitting: Port-split technology supports up to 128× 200Gb/s InfiniBand connections for dense top-of-rack and cluster deployments.
  • Integrated management: NVIDIA MLNX-OS and the onboard subnet manager provide out-of-box fabric bringup and switch management for deployments up to 2,000 nodes.
  • Serviceable rack design: Dual hot-swappable power supplies and 6+1 hot-swappable fan modules support reliable operation and easier maintenance.

TECHNICAL DETAILS

Built into the system

64× 400G NDR InfiniBand Fabric

The QM9700 delivers 64 non-blocking 400Gb/s NDR InfiniBand ports with up to 51.2Tb/s of aggregate bidirectional throughput. It is designed for AI clusters, HPC environments, research systems, and other workloads that require extremely fast node-to-node communication.

Dense 1U OSFP Design

Thirty-two OSFP connectors provide the physical interface for the switch’s 64 400G links. This high-density design brings substantial fabric capacity into one 1U rackmount chassis, helping reduce rack space and network complexity.

Flexible 200G Scaling

NVIDIA port-split technology allows the QM9700 to support up to 128× 200Gb/s InfiniBand connections. This provides a path for connecting high-density compute environments while maintaining a compact network architecture.

In-Network Computing with SHARPv3

NVIDIA SHARPv3 enables collective operations to be processed within the InfiniBand fabric rather than only at endpoint systems. This reduces data movement across the network and can improve performance for distributed AI training, simulations, and analytics workloads.

Intelligent Fabric Performance

RDMA, adaptive routing, congestion control, quality of service, and enhanced virtual-lane mapping support efficient, low-latency communication across the fabric. NVIDIA InfiniBand also includes self-healing capabilities designed to help maintain availability in large-scale deployments.

Managed 1U Infrastructure Platform

The MQM9700-NS2F includes NVIDIA MLNX-OS, an onboard subnet manager, dual hot-swappable power supplies, and 6+1 hot-swappable fan units. It measures 1U high, 17.0 inches wide, and 26.0 inches deep, with forward power-to-connector airflow.

No video is needed for this one. The product and port-diagram sections will communicate the value better than a generic manufacturer video.