Q3400-RA (P/N: 920-9B36F-00RX-8S0) Detailed Product Overview + Full Specification Table
Product Overview (Professional & Detailed Version)
The NVIDIA Quantum Q3400-RA is an enterprise‑class managed InfiniBand switch powered by the cutting‑edge Quantum‑3 (Quantum‑X800) ASIC, purpose‑built for the most demanding AI training clusters, high‑performance computing (HPC), large‑scale data centers, and next‑generation cloud infrastructures. It adopts industry‑leading 200Gb/s per lane SerDes technology to deliver breakthrough 800Gb/s XDR InfiniBand bandwidth, with a non‑blocking switching capacity of up to 115.2 Tbps to fully unlock the performance of GPU clusters and high‑speed computing nodes.
This switch provides 72 high‑speed OSFP ports, supporting 144 logical 800G XDR ports while maintaining full backward compatibility with 400G NDR, 200G HDR, 100G EDR, 56G FDR, and 40G QDR InfiniBand rates, protecting existing network investments. Integrated with SHARP v4 in‑network computing acceleration, it offloads collective communication operations to the switch fabric, drastically reducing CPU overhead, lowering end‑to‑end latency, and significantly accelerating AI model training, large‑scale parallel computing, and real‑time data processing workloads.
Built with a standard 4U rack‑mount, air‑cooled chassis and C2P (Connector‑to‑Power) airflow, the Q3400-RA supports 8 hot‑swappable AC power supplies with full N+N redundancy to ensure uninterrupted operation in mission‑critical environments. It offers comprehensive management including CLI, Web UI, SNMP, gNMI, Redfish, and NVIDIA UFM for centralized orchestration, real‑time telemetry, intelligent fault detection, and automated O&M.
With enterprise‑grade security (secure boot, IRoT, hardware‑based encryption), adaptive routing, advanced congestion control, and high‑radix scalable topology, the Q3400-RA supports two‑level fat‑tree architecture connecting up to 10,368 NICs to satisfy ultra‑large‑scale cluster expansion. It is the ideal core switching solution for AI supercomputing centers, large language model (LLM) training, scientific research, financial computing, and cloud data centers requiring extreme bandwidth, ultra‑low latency, and carrier‑grade reliability.
Full Specifications Table
| Parameter |
Details |
| Model |
NVIDIA Quantum Q3400-RA |
| Part Number |
920-9B36F-00RX-8S0 |
| Switch Chip |
NVIDIA Quantum-3 (Quantum-X800) |
| Port Configuration |
72 × OSFP (supports 144 × 800Gb/s logical ports) |
| Transmission Speed |
800G XDR / 400G NDR / 200G HDR / 100G EDR / 56G FDR / 40G QDR |
| Switching Capacity |
115.2 Tbps non-blocking |
| Management CPU |
Intel i3-8100H 4-core 3.0GHz |
| System OS |
NVIDIA NVOS |
| Management Interfaces |
1 × 1GbE RJ45 management port, 1 × USB, 1 × RS232 serial |
| Management Protocols |
CLI, Web, SNMP v2/v3, gNMI, Redfish, REST API, UFM, NetQ |
| Key Technologies |
SHARP v4, Adaptive Routing, Adaptive Congestion Control, QoS, Subnet Management (SM) |
| Security |
Secure Boot, IRoT, CPLD/Switch IC security, access authentication, encryption |
| Form Factor |
4U, 19-inch rack-mountable |
| Dimensions |
177.8mm (H) × 438mm (W) × 850mm (D) |
| Cooling |
Air-cooled, C2P (front-to-back) airflow, redundant hot-swappable fans |
| Power Supply |
8 × hot-swappable AC PSU (200–240V AC), N+N redundancy, power cords not included |
| Operating Temperature |
0°C ~ 40°C |
| Storage Temperature |
-20°C ~ 70°C |
| Humidity |
10% ~ 90% non-condensing |
| Application Scenarios |
AI training clusters, HPC, data center spine/leaf, GPU interconnection, supercomputing |