NVIDIA Quantum‑2 MQM9790‑NS2R
64 ports of 400Gb/s NDR InfiniBand in 1U — 51.2 Tb/s aggregate throughput · 66.5 BPPS · SHARPv3 in‑network computing · C2P airflow
Switch radix
64 x 400Gb/s
Bisection throughput
51.2 Tb/s
Packet capacity
66.5 BPPS
Hong Kong Starsurge Group Co., Limited
Hong Kong Starsurge Group Co., Limited is a technology-driven provider of network hardware, IT services, and system integration solutions. Founded in 2008, the company serves customers worldwide with products including network switches, NICs, wireless access points, controllers, cables, and related networking equipment. Backed by an experienced sales and technical team, Starsurge supports industries such as government, healthcare, manufacturing, education, finance, and enterprise. The company also offers IoT solutions, network management systems, custom software development, multilingual support, and global delivery. With a customer-first approach, Starsurge focuses on reliable quality, responsive service, and tailored solutions that help clients build efficient, scalable, and dependable network infrastructure.
Product overview
The NVIDIA Quantum‑2 MQM9790‑NS2R is an unmanaged 1U switch offering 64 ports of 400Gb/s NDR InfiniBand. It delivers an aggregate bidirectional throughput of 51.2 Tb/s and more than 66.5 billion packets per second. Designed for extreme‑scale AI and HPC, it integrates third‑generation NVIDIA SHARP (SHARPv3) in‑network computing, boosting AI acceleration up to 32× compared to prior generation. This model features connector‑to‑power (C2P) airflow, is ready for NVIDIA UFM advanced fabric management, and supports flexible topologies (Fat Tree, DragonFly+, Torus, SlimFly).
Unmatched performance
64 non‑blocking 400Gb/s ports, 51.2 Tb/s aggregated throughput, 66.5 BPPS. Port‑split technology enables 128 ports of 200Gb/s for double‑density top‑of‑rack deployments.
In‑network computing SHARPv3
Third‑gen NVIDIA SHARP technology accelerates collective communications. Data aggregation happens inside the fabric, reducing traffic and delivering up to 32× higher AI acceleration than previous generation.
NVIDIA SHARPv3 technology
The MQM9790‑NS2R embeds the third generation of NVIDIA Scalable Hierarchical Aggregation and Reduction Protocol (SHARP). By offloading data reduction operations (all‑reduce, broadcast, etc.) to the switch fabric, SHARPv3 dramatically cuts inter‑node traffic and reduces application runtime. This in‑network computing capability is essential for massively parallel AI training and complex HPC simulations.
Typical deployments
- Large‑scale AI supercomputers (LLM training, simulation)
- High‑frequency financial modeling and risk analysis
- Multi‑tenant cloud HPC with isolated fabric partitions
- Research labs with hybrid Fat‑Tree / DragonFly+ topologies
- Enterprise core for data‑intensive workloads (genomics, climate)
Compatibility
- Backward compatible with 200Gb HDR, 100Gb EDR, and previous InfiniBand generations
- Interoperable with NVIDIA ConnectX‑6, ConnectX‑7, and BlueField‑3 DPUs
- OSFP connectors: passive/active copper, active fiber, optical modules
- MLNX‑OS software suite (used internally); UFM management ready
Technical specifications
| Parameter | Details |
| Ports & data rate | 64 ports 400Gb/s NDR (32x OSFP connectors) – non‑blocking |
| Aggregate throughput | 51.2 Tb/s bidirectional |
| Packet rate | >66.5 billion packets/sec |
| Port split mode | 128 ports of 200Gb/s using NVIDIA port‑split technology |
| Form factor / dimensions | 1U rack mount: 1.7 x 17.0 x 26.0 in (43.6 x 438 x 660.4 mm) |
| Weight | 14.5 kg |
| Power supply | 1+1 redundant, hot‑swappable; input 200‑240Vac; 80+ Gold, ENERGY STAR |
| Cooling | Connector‑to‑power (C2P / reverse airflow); hot‑swap fans |
| CPU / memory / storage | x86 Coffee Lake i3; 8GB DDR4 SO‑DIMM (2666 MT/s); 16GB M.2 SATA SSD |
| Management ports | 1x USB 3.0, 1x USB‑I2C, 1x RJ45, 1x RJ45 (UART) |
| Software | MLNX‑OS (unmanaged, UFM‑ready) |
| Operating conditions | 0‑40°C, 10‑85% non‑condensing, altitude up to 3050m |
| Safety / compliance | CE, FCC, VCCI, ICES, RCM; RoHS, CB, cTUVus, CE, CU |
| Warranty | 1 year (extendable, please confirm) |
Selection guide
| OPN | Description |
| MQM9790-NS2R | Unmanaged, 64x400G, C2P airflow (rear‑to‑front) |
Buyer checklist
- ✔ Confirm airflow direction C2P matches your data center hot/cold aisle.
- ✔ Verify power input (200‑240Vac) and redundancy requirements.
- ✔ This is an unmanaged switch; plan to use NVIDIA UFM for advanced management.
- ✔ For 128 ports of 200G, ensure your cables/optics support split‑mode.
- ✔ Check compatibility with existing HDR/EDR infrastructure.
Why choose Quantum‑2
Highest radix 1U 400G switch with 51.2Tb/s capacity. SHARPv3 in‑network computing reduces job completion time. Self‑healing network, QoS, adaptive routing, and enhanced VL mapping. Energy‑efficient design lowers OPEX.
Service & support
1‑year limited hardware warranty (extendable). Advanced replacement available. 24/7 technical support through Hong Kong Starsurge Group. Firmware updates and UFM licenses available. Please contact our sales team for exact terms.
Frequently asked questions
Can I use MQM9790‑NS2R with 200Gb/s optics? Yes, using NVIDIA port‑split technology each OSFP port can operate as 2x200Gb/s, giving 128 ports of 200Gb/s.
Does this switch support RDMA? Absolutely. NDR InfiniBand natively supports RDMA, plus adaptive routing and congestion control for lowest latency.
How is MQM9790‑NS2R managed without on‑board subnet manager? It is designed to be used with NVIDIA UFM (Unified Fabric Manager) for comprehensive monitoring and management.
Is it backward compatible with ConnectX‑5 (EDR)? Yes, the switch supports mixed‑rate links and auto‑negotiation down to EDR (100Gb/s), though for optimal performance we recommend ConnectX‑6 or newer.
Important notes & precautions
- Install only in a rack with adequate airflow – ensure fan trays match your C2P direction.
- Use only supported OSFP modules (passive/active copper or optical) listed in the compatibility matrix.
- Power input must be 200‑240Vac; not compatible with 110V.
- Operating altitude up to 3050m; derating may apply above 1800m.
- Some specifications (e.g., exact CPU model, SSD size) are typical but may change; confirm with order.
Compatibility matrix
| Component | Compatible with MQM9790‑NS2R |
| NVIDIA ConnectX‑7 (NDR) | ✅ 400Gb/s native |
| NVIDIA ConnectX‑6 (HDR) | ✅ 200Gb/s (using split or 1x200G) |
| NVIDIA BlueField‑3 DPU | ✅ Full NDR support |
| Legacy QDR / FDR | ❌ Not directly; requires conversion |
| MLNX‑OS / UFM versions | ✅ Latest UFM 6.10+ recommended |
Related products
- ▪ NVIDIA Quantum‑2 MQM9700 (managed version)
- ▪ NVIDIA ConnectX‑7 NDR adapter
- ▪ MCP1600 OSFP/AOC cables (400G)
- ▪ NVIDIA UFM Cyber‑AI platform
Related guides / comparisons
- ▪ MQM9790 vs. MQM9700: UFM‑ready vs. on‑board subnet manager
- ▪ 400G NDR vs. 200G HDR trade‑offs
- ▪ Building a DragonFly+ with Quantum‑2
- ▪ SHARPv3 performance white paper
Key facts
- ▪ 64 ports 400Gb/s NDR InfiniBand
- ▪ 51.2 Tb/s aggregate throughput
- ▪ 66.5 billion packets/sec
- ▪ 128 ports 200Gb/s via port split
- ▪ SHARPv3 in‑network computing
- ▪ UFM ready (unmanaged)
- ▪ C2P reverse airflow
- ▪ 32 OSFP connectors
Suggested schema fields (Product/ProductGroup):
brand, model, mpn (MQM9790-NS2R), sku, offers, availability, manufacturer (NVIDIA), isAccessoryOrSparePartFor (cables/optics), hasVariant, productID, description, weight, dimensions, category, warranty.
