NVIDIA Quantum-3 Q3400-RA XDR 800Gb/s InfiniBand Switch System
The NVIDIA Q3400-RA is a high-performance, 4U rack-mounted InfiniBand switch system engineered for next-generation AI and HPC data centers. Built on the groundbreaking NVIDIA Quantum-3 ASIC, this network switch delivers an industry-leading 115.2 Tb/s aggregate throughput through 144 non-blocking 800Gb/s ports. It is designed for massive scalability, ultra-low latency, and exceptional power efficiency, making it the ideal spine solution for large-scale clusters and AI training fabrics.
Key Selling Points
- Unmatched Performance: 144 ports of 800Gb/s InfiniBand for a total bandwidth of 115.2 Tb/s.
- Quantum-3 ASIC: Leverages 200Gb/s SerDes technology for superior radix and efficiency.
- Advanced Management: Features a dedicated Fabric Network Management (FNM) port for streamlined control via NVIDIA UFM.
- Robust Design: 4U form factor with redundant, hot-swappable power supplies and cooling for maximum reliability.
- Future-Proof: Supports optional router functionality for multi-site cluster expansion and a wide range of network topologies.
Characteristics
- Fixed-configuration 4U InfiniBand switch system
- 144 XDR 800Gb/s ports over 72 OSFP cages
- Air-cooled with front-to-rear (C2P) airflow
- 8 hot-swappable, redundant power supplies (200-240V AC)
- 10 hot-swappable fan units
- Dedicated management ports: FNM (OSFP), MGT (RJ45), Console (RJ45), USB
- Support for NVIDIA SHARP™, adaptive routing, and performance isolation
Technology
The Q3400-RA network switch is powered by the NVIDIA Quantum-3 switch ASIC, utilizing cutting-edge 200Gb/s-per-lane SerDes technology. It supports the XDR InfiniBand standard, which operates at 200Gb/s per lane with 64b/66b encoding for an effective data rate of 800Gb/s per port. The system also supports backward compatibility with NDR (400Gb/s), HDR (200Gb/s), EDR (100Gb/s), FDR (56Gb/s), and QDR (40Gb/s) speeds. It incorporates advanced features like NVIDIA's Scalable Hierarchical Aggregation and Reduction Protocol (SHARP) for accelerated collective operations.
Working Principle
The SN 2000 series switch operates as a high-radix, non-blocking crossbar switch. Data packets enter through the OSFP cages on the front panel. The Quantum-3 ASIC performs packet switching, routing, and advanced processing (like SHARP offloads) at line rate. The managed system runs NVOS (NVIDIA Operating System), which provides full control over the switching fabric, configuration, and monitoring via in-band or out-of-band management interfaces. The separate FNM port provides a dedicated path for fabric management traffic, ensuring data ports are fully utilized for compute workloads.
Application & Usage
The NVIDIA SN 2000 Q3400-RA is engineered for the most demanding high-performance computing and artificial intelligence environments. Its primary applications include:
- AI Training Clusters: Connecting thousands of GPUs for large-scale model training.
- High-Performance Computing (HPC): Serving as a high-bandwidth spine in fat-tree or dragonfly+ topologies for scientific simulations.
- Hyper-Converged Infrastructure (HCI): Providing the low-latency, high-throughput backbone for data-intensive applications.
- Cloud Data Centers: Building efficient, scalable fabrics for cloud service providers.
- Multi-Site Clusters: Utilizing its optional routing capabilities to connect clusters across different geographical locations.
Specification Table
| Attribute | Specification |
| Model | Q3400-RA |
| Form Factor | 4U Rackmount |
| Switch ASIC | NVIDIA Quantum-3 |
| Data Ports | 144 x XDR 800Gb/s InfiniBand |
| Hardware Ports | 72 x OSFP Cages |
| Max Throughput | 115.2 Tb/s |
| Management Ports | FNM (OSFP), 2x MGT (RJ45), Console (RJ45), USB 2.0 |
| Power Supplies | 8 Hot-Swappable, 2000W (200-240V AC) |
| Cooling | 10 Hot-Swappable Fans, C2P Airflow |
| Typical Power | ~2.9 kW (with passive cables) |
| Dimensions (HxWxD) | 7" x 17.24" x 33.5" (177.8mm x 438mm x 850mm) |
| Weight | ~60 kg (with 8 PSUs) |
Advantages & Selling Points
- Industry-Leading Density & Performance: Offers the highest port density and throughput in a 4U form factor, reducing data center footprint and cost per port.
- Uncompromising Reliability: Features full hardware redundancy (PSUs, fans) and robust thermal management for 24/7 operation in demanding environments.
- Investment Protection: Backward compatibility with previous InfiniBand generations (NDR, HDR, EDR) protects existing infrastructure investments.
- Enhanced AI & HPC Capabilities: Native support for NVIDIA SHARP dramatically accelerates AI and HPC workloads by offloading collective operations from the CPU.
- Simplified Fabric Management: The dedicated FNM port, working with NVIDIA UFM, provides unparalleled visibility and control without consuming data ports.
Service & Support
We provide comprehensive support for all our products, including the network switch Q3400-RA. Our services encompass:
- Warranty: Standard manufacturer warranty with options for extended coverage.
- Availability: Large in-stock inventory ensures quick shipment and short lead times.
- Technical Support: 24/7 customer consultation and technical support from our experienced team.
- Global Logistics: Reliable worldwide shipping with full tracking and customs support.
Frequently Asked Questions (FAQ)
Q: What is the difference between the Q3200-RA and Q3400-RA models?
A: The Q3400-RA is a single, large 4U switch with 144 ports. The Q3200-RA is a 2U chassis containing two independent 36-port switches, totaling 72 ports, ideal for integrating with existing NDR/HDR infrastructures.
Q: What types of transceivers and cables are compatible with this switch?
A: The system uses OSFP connectors and supports a wide range of official NVIDIA and certified third-party optical transceivers and DAC/AOC cables for speeds from 40G to 800G. Unused ports must be covered with provided thermal caps.
Q: Can this switch be used in a mixed InfiniBand environment?
A: Yes. The SN 2000 Q3400-RA supports auto-negotiation and is backward compatible with NDR, HDR, EDR, FDR, and QDR InfiniBand devices.
Q: How is the switch managed?
A: It is managed by NVIDIA's NVOS operating system, which can be accessed via the dedicated Ethernet management ports (MGT), the serial console, or the Fabric Network Management (FNM) port for integration with NVIDIA UFM.
Q: What is the power redundancy configuration?
A: The system has 8 power supply slots. It requires a minimum of 2 PSUs for operation but is designed for N+N redundancy. All PSUs are hot-swappable.
Precautions
- Installation: Must be installed horizontally in a standard 19" rack. Ensure at least 8cm (3 inches) of clearance around ventilation openings.
- Environment: Designed for controlled data center environments. Operating temperature: 0°C to 35°C (with C2P airflow).
- Safety: The unit is heavy (~60kg). Use a mechanical lift for installation. Disconnect all power cords before servicing.
- Electromagnetic Compatibility (EMC): Complies with CE, FCC, VCCI, ICES, and RCM standards.
- Compatibility: Use only NVIDIA-supplied or certified cables for management interfaces (e.g., console cable) to prevent potential I²C communication issues.
Company Introduction
With over a decade of industry experience, we have established ourselves as a leading provider of premium networking solutions. Our extensive facilities and strong technical team support a vast inventory, enabling us to offer competitive pricing and superior service on a global scale.
We are authorized partners for top-tier brands including NVIDIA Networking (Mellanox), Ruckus, Aruba, and Extreme. Our core product range encompasses brand-new original equipment such as network switches, adapters, wireless access points, controllers, and cabling solutions.
