 
|
NVIDIA Quantum-2 QM9790-NS2F 400Gb/s InfiniBand Switch – 64-Port NDR Smart Switch with 51.2 Tb/s Throughput
Product Details:
| Brand Name: | Mellanox |
| Model Number: | MQM9790-NS2F(920-9B210-00FN-0D0) |
| Document: | MQM9700 series.pdf |
Payment & Shipping Terms:
| Minimum Order Quantity: | 1pcs |
|---|---|
| Price: | Negotiate |
| Packaging Details: | outer box |
| Delivery Time: | Based on inventory |
| Payment Terms: | T/T |
| Supply Ability: | Supply by project/batch |
|
Detail Information |
|||
| Application: | Server | Condition: | New And Original |
|---|---|---|---|
| Aggregate Bandwidth: | 51.2 Tb/s Bidirectional | Management: | MLNX-OS With CLI, WebUI, SNMP, JSON Support |
| Name: | MQM9790-NS2F InfiniBand Mellanox Network Switch Scaling Out Data Centers With NDR 400G InfiniBand Smart Switches | Keyword: | Mellanox 400g Switch |
| Highlight: | NVIDIA Quantum-2 400Gb/s InfiniBand Switch,64-Port NDR Smart Switch,51.2 Tb/s Throughput Network Switch |
||
Product Description
NVIDIA Quantum-2 QM9790-NS2F 400Gb/s InfiniBand Switch
64-port NDR managed smart switch with 51.2 Tb/s aggregate throughput and SHARPv3 in-network computing capabilities.
Product Overview
The NVIDIA Quantum-2 QM9790-NS2F delivers 64 ports of 400Gb/s InfiniBand in a 1U chassis, enabling extreme-scale AI and HPC clusters with industry-leading density, ultra-low latency, and self-healing fabric. Designed for external management with NVIDIA UFM, it streamlines data center operations while supporting Fat Tree, DragonFly+, and Torus topologies.
This unmanaged (externally managed) 400Gb/s InfiniBand fixed-configuration switch provides 32 OSFP ports, each capable of 400Gb/s (or 128 ports at 200Gb/s with port-split technology), and offers an aggregate bidirectional throughput of 51.2 Tb/s. With support for NVIDIA SHARPv3, RDMA, adaptive routing, and advanced congestion control, the QM9790 is built for the most demanding HPC, AI, and enterprise workloads.
Key Features
- 64 non-blocking 400Gb/s ports (32 OSFP connectors)
- 51.2 Tb/s aggregate bidirectional throughput
- Port-split technology: up to 128 ports @ 200Gb/s
- SHARPv3 in-network computing - 32× AI acceleration
- Sub-microsecond latency with adaptive routing
- Self-healing fabric, QoS, and congestion control
- 1+1 redundant hot-swappable power (200-240Vac)
- Front-to-rear (P2C) airflow, hot-swappable fans
- External management via NVIDIA UFM / MLNX-OS
- Backward compatible with previous InfiniBand generations
Technology: NVIDIA Quantum-2 Architecture
The Quantum-2 platform integrates the third-generation SHARP technology (SHARPv3), enabling in-network data aggregation that reduces data movement and accelerates AI training and HPC simulations. Combined with Remote Direct Memory Access (RDMA) and NVIDIA's Scalable Hierarchical Aggregation and Reduction Protocol, the QM9790-NS2F dramatically improves application performance while data traverses the fabric.
Typical Deployments
- AI & Machine Learning clusters - large-scale GPU-accelerated training with SHARPv3
- High-performance computing (HPC) - research simulations, weather modeling, and genomics
- Data center spine-leaf - 400G InfiniBand fabrics for low-latency storage and compute
- Enterprise & cloud - scalable, multi-tenant networks with UFM management
- Government & education - research labs and supercomputing centres
Compatibility
| Component / Ecosystem | Support |
|---|---|
| InfiniBand Speed | NDR 400Gb/s, backward compatible with HDR, EDR, FDR |
| Cabling | Passive/active copper or active fiber; OSFP connectors |
| Management | NVIDIA UFM, MLNX-OS, CLI, WebUI, SNMP, JSON-RPC |
| Topologies | Fat Tree, SlimFly, DragonFly+, multi-dimensional Torus |
| Operating Systems | Linux, Windows (with InfiniBand stack), VMware |
| GPUs / Accelerators | NVIDIA HGX, DGX, and third-party GPU servers |
Specifications
| Parameter | QM9790-NS2F |
|---|---|
| Ports / Speed | 64 ports 400Gb/s (32 OSFP); up to 128 ports @ 200Gb/s with split |
| Aggregate throughput | 51.2 Tb/s bidirectional |
| Packet forwarding | >66.5 billion packets per second (BPPS) |
| Form factor | 1U rack-mount, 43.6 x 438 x 660.4 mm |
| Weight | 14.5 kg |
| Power supply | 1+1 redundant, hot-swappable, 200-240Vac, 80+ Gold, ENERGY STAR |
| Cooling | Front-to-rear (P2C) airflow, hot-swappable fans |
| Management ports | 1x USB 3.0, 1x USB (I2C), 1x RJ45, 1x RJ45 (UART) |
| CPU / Memory | x86 Coffee Lake i3, 8GB DDR4 SO-DIMM, 16GB M.2 SSD |
| Software | MLNX-OS (external management via UFM) |
| Operating temperature | 0°C to 40°C |
| Humidity (operating) | 10% - 85% non-condensing |
| Altitude | Up to 3050 m |
| Safety / EMC | RoHS, CB, cTUVus, CE, CU; FCC, VCCI, ICES, RCM |
| Warranty | 1 year (extended options available) |
Selection Guide
| Orderable Part Number | Description | Airflow | Management |
|---|---|---|---|
| MQM9790-NS2F | 64-port 400Gb/s, 32 OSFP, unmanaged, P2C airflow | Front-to-rear (P2C) | External (UFM) |
| MQM9790-NS2R | 64-port 400Gb/s, 32 OSFP, unmanaged, C2P airflow | Connector-to-power (C2P) | External (UFM) |
| MQM9700-NS2F | 64-port 400Gb/s, managed, P2C airflow | Front-to-rear | On-board subnet manager |
| MQM9700-NS2R | 64-port 400Gb/s, managed, C2P airflow | Connector-to-power | On-board subnet manager |
Choosing between QM9700 (managed) and QM9790 (externally managed) depends on cluster size and management preference. The QM9790-NS2F is recommended for large-scale deployments using NVIDIA UFM for advanced fabric optimization.
Advantages
- Unmatched density: 64 ports of 400G in 1U - reduces rack space and power per port.
- In-network acceleration: SHARPv3 reduces data movement and accelerates AI/ML workloads up to 32×.
- Flexible topologies: Supports Fat Tree, DragonFly+, Torus, and more for any cluster design.
- Operational efficiency: External UFM management provides proactive monitoring, provisioning, and troubleshooting.
- Investment protection: Backward compatible with HDR/EDR and future-proof for NDR200/400.
Service & Support
Hong Kong Starsurge Group provides global delivery, multilingual technical support, and customized integration services for the NVIDIA Quantum-2 series. Our team offers pre-sales consultation, rack integration, cabling design, and post-warranty support. Extended warranties and on-site replacement options are available upon request.
Frequently Asked Questions
What is the difference between QM9700 and QM9790?
QM9700 includes an on-board subnet manager and is fully managed out-of-the-box for up to 2000 nodes. QM9790 is externally managed and leverages NVIDIA UFM for advanced fabric management, ideal for larger, more dynamic environments.
QM9700 includes an on-board subnet manager and is fully managed out-of-the-box for up to 2000 nodes. QM9790 is externally managed and leverages NVIDIA UFM for advanced fabric management, ideal for larger, more dynamic environments.
Can I use QM9790-NS2F with existing HDR InfiniBand infrastructure?
Yes, the switch is backward compatible with HDR, EDR, and FDR speeds, allowing incremental upgrades.
Yes, the switch is backward compatible with HDR, EDR, and FDR speeds, allowing incremental upgrades.
What cabling is supported?
OSFP connectors support passive/active copper and active fiber cables. Both optical modules and direct-attach cables can be used.
OSFP connectors support passive/active copper and active fiber cables. Both optical modules and direct-attach cables can be used.
Does the switch support 200Gb/s operation?
Yes, using NVIDIA port-split technology, the QM9790 can provide up to 128 ports at 200Gb/s.
Yes, using NVIDIA port-split technology, the QM9790 can provide up to 128 ports at 200Gb/s.
What is the typical power consumption?
Please confirm with your sales representative; typical values depend on configuration and cabling.
Please confirm with your sales representative; typical values depend on configuration and cabling.
Precautions
- Ensure the operating ambient temperature remains between 0°C and 40°C.
- Use only NVIDIA-qualified optical modules and cables for optimal performance.
- Proper airflow direction (P2C or C2P) must match the installation environment.
- Do not block fan vents; maintain at least 5 cm clearance on both sides.
- Power supply must meet 200-240Vac requirements; grounding is mandatory.
About Hong Kong Starsurge Group
Founded in 2008, Hong Kong Starsurge Group Co., Limited is a technology-driven provider of network hardware, IT services, and system integration solutions. With an experienced sales and technical team, we serve industries including government, healthcare, manufacturing, education, finance, and enterprise. Starsurge offers IoT solutions, network management systems, custom software development, multilingual support, and global delivery.
Partner with Starsurge - we provide competitive pricing, fast shipping, and dedicated engineering support for NVIDIA, Mellanox, and other leading networking brands.
Key Facts
| Ports | 64 x 400Gb/s (32 OSFP) |
| Throughput | 51.2 Tb/s |
| Packet rate | >66.5 BPPS |
| Management | External (UFM) |
| Airflow | P2C (Front-to-rear) |
| Warranty | 1 year |
Compatibility Matrix
| Category | Compatible Products / Standards |
|---|---|
| GPU Servers | NVIDIA DGX-2, DGX A100, HGX-based servers, third-party GPU nodes |
| Storage | NVMe over InfiniBand, Lustre, WekaIO, IBM Spectrum Scale |
| Management | NVIDIA UFM, MLNX-OS, OpenSM, Prometheus/Grafana (via UFM) |
| Cable types | OSFP to OSFP, OSFP to QSFP (with adapter), active optical, DAC |
Buyer Checklist
- Confirm airflow direction (P2C vs. C2P) matches your data center hot-aisle/cold-aisle layout.
- Plan for external management - ensure UFM licenses and server infrastructure are ready.
- Validate power input (200-240Vac) and redundancy requirements.
- Check OSFP cable lengths and types (copper/fiber) for your rack distances.
- Request extended warranty or on-site support if needed for mission-critical deployment.
Want to Know more details about this product