Specifications
Orderable Part Number
MQM9700-NS2R
NVIDIA SKU
920-9B210-00RN-0M2
System
NVIDIA Quantum-2 QM9700-based NDR InfiniBand switch
Ports
64 × NDR 400 Gb/s on 32 OSFP connectors
Port Split
Up to 128 × 200 Gb/s ports
Management
Managed (on-board subnet manager, MLNX-OS)
Airflow
Connector-to-Power (C2P), reverse
Switching Capacity
25.6 Tb/s unidirectional / 51.2 Tb/s bidirectional
Packet Rate
> 66.5 billion packets per second (BPPS)
Latency
130 ns
Switch IC
NVIDIA Quantum-2
In-Network Computing
NVIDIA SHARP / SHARPv3 (NCCL collective offload)
Routing
Adaptive routing with congestion control; RDMA
Topologies
Fat Tree, DragonFly+, SlimFly, multi-dimensional Torus
CPU / Memory / Boot
x86 Coffee Lake i3, 8 GB DDR4 SO-DIMM (2666 MT/s), 16 GB M.2 SSD
Management Ports
1x USB 3.0, 1x USB (I2C), 1x RJ45 Ethernet, 1x RJ45 UART
Power Supply
1+1 redundant, hot-swappable; 200–240 VAC; 80 PLUS Gold+
PSU Accessory
MTEF-PSF-AC-J, 2,000 W AC PSU, P2C airflow
Typical Power
747 W
Form Factor
1U rack mount, standard depth
Dimensions
43.6 mm H × 433.2 mm W × 590.6 mm D
Operating Temperature
0 °C to 40 °C (forward airflow)
Backward Compatibility
Interoperates with prior InfiniBand generations
Overview
The NVIDIA Quantum-2 QM9700 MQM9700-NS2R is a fully managed 1U InfiniBand switch carrying 64 ports of NDR 400 Gb/s on only 32 physical OSFP connectors. A single switch delivers 51.2 Tb/s of bidirectional throughput with more than 66.5 billion packets per second and 130 ns port-to-port latency — the radix that makes a full non-blocking fabric for a large GPU pod possible without a three-tier tree.
This is the managed model (part number MQM9700-NS2R, NVIDIA SKU 920-9B210-00RN-0M2) with an on-board subnet manager running MLNX-OS and connector-to-power (reverse) airflow. The QM9700 platform brings SHARPv3 in-network collective offload, so NCCL all-reduce traffic is aggregated in the switch fabric rather than in the GPUs.
QS Compute supplies the Quantum-2 QM97XX family — QM9700 managed, QM9790 externally managed and the QM9701 48 VDC DGX switch — together with OSFP transceivers, DAC/AOC cables and rail kits.
Key Benefits
64 NDR 400 Gb/s ports in 1U — high radix per rack unit, fewer fabric tiers; 51.2 Tb/s with 130 ns latency for tightly-coupled GPU training; SHARPv3 in-network computing offloads NCCL collectives from the GPU cores; port-split to 128 × 200 Gb/s for denser, lower-bandwidth topologies; 1+1 redundant hot-swap PSUs keep a training job alive through a PSU fault.
Applications
Large-scale AI/ML training clusters; HPC and GPU supercomputing fabrics; storage and compute convergence over RDMA; research and national-lab clusters; high-radix data-centre spine for NDR InfiniBand deployments.
Request a Quote — NVIDIA QUANTUM-2 QM9700 MQM9700-NS2R — 64-PORT NDR 400G INFINIBAND SWITCH
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →