 
|
NVIDIA Q3200-RA 72-Port 400Gb/s NDR InfiniBand Switch Dual System
Dettagli:
| Marca: | Mellanox |
| Numero di modello: | Q3200-RA (920-9B34F-00RX-FS0) |
| Documento: | q3200-ra-q3400-ra-xdr-800gb...al.pdf |
Termini di pagamento e spedizione:
| Quantità di ordine minimo: | 1 pezzo |
|---|---|
| Prezzo: | Negotiate |
| Imballaggi particolari: | Scatola esterna |
| Tempi di consegna: | Basato sull'inventario |
| Termini di pagamento: | T/T |
| Capacità di alimentazione: | Fornitura per progetto/batch |
|
Informazioni dettagliate |
|||
| Modello n.: | Q3200-RA (920-9B34F-00RX-FS0) | velocità di trasmissione: | 800 g |
|---|---|---|---|
| Porti: | 36 | Funzione: | LACP, stackable, supporto VLAN |
| Tecnologia: | Infiniband | Velocità massima: | XDR |
| Pacchetto di trasporto: | Imballaggio | Specifica: | 43.6*433.2*590,6 mm |
| Marchio: | Mellanox | ||
| Evidenziare: | NVIDIA InfiniBand switch 400Gb/s,Mellanox NDR switch 72-port,Dual system network switch Q3200-RA |
||
Descrizione di prodotto
NVIDIA Q3200-RA 72-Port 400Gb/s NDR InfiniBand Switch Dual System
Q3200-RA - 72-Port NDR 400Gb/s InfiniBand Dual-System Switch
Engineered for high-density AI and HPC clusters. The Q3200-RA integrates two independent 36-port Quantum-3 switches in a compact 2U air-cooled chassis.
28.8 Tb/s Total System Throughput
72 x 400Gb/s Ports (2 x 36 independent switches)
ASIC Technology: 2x NVIDIA Quantum-3 (NDR 400Gb/s)
Form Factor: 2U Rackmount • Air-cooled • C2P Airflow
Redundancy: 6x Hot-Swappable PSUs • 12x Fans
Product Overview
The NVIDIA Quantum-3 Q3200-RA is a dual-system InfiniBand switch platform, packing two independent 36-port NDR switches into a single 2U chassis. Leveraging NVIDIA's Quantum-3 ASIC, it delivers 400Gb/s per port with ultra-low latency. Ideal for building cost-effective leaf-spine fabrics, the Q3200-RA enables organizations to scale AI infrastructures without sacrificing port density or performance. Each of the two switch planes operates independently, offering physical redundancy in a compact form factor.
Key Features
- Dual-System Architecture: Two independent 36-port NDR 400Gb/s switches in a single 2U chassis
- High Throughput: 28.8 Tb/s aggregate system switching capacity
- In-Network Computing: Full support for NVIDIA SHARPv3 (Scalable Hierarchical Aggregation and Reduction Protocol)
- Advanced Fabric Management: Dedicated Fabric Management Port (FNM) for NVIDIA UFM telemetry and AI-driven orchestration
- Reliability: Redundant, hot-swappable 6 PSUs and 12 fan units with Connector-to-Power (C2P) airflow
Technology & Performance
Each of the two switch modules inside the Q3200-RA is built on the NVIDIA Quantum-3 Switch ASIC, utilizing low-latency 50Gb/s SerDes (4 lanes per port). The switch features adaptive routing to avoid hot spots, congestion control algorithms, and performance isolation for multi-tenant AI clouds. It supports GPU-initiated communication and NVIDIA GPUDirect RDMA for accelerated compute jobs.
Optimized Topologies: Fat Tree (Leaf-Spine), Slim Fly, Dragonfly+
Compatibility Matrix
| Component Type | Recommended Models |
|---|---|
| Adapter Cards | NVIDIA ConnectX-7 (NDR 400Gb/s), ConnectX-6 (HDR 200Gb/s) |
| Cables (Passive) | OSFP to OSFP DAC, OSFP to 2x OSFP Splitter |
| Cables (Active) | OSFP Active Optical Cables (AOC), MPO Breakout Cables |
| Management Software | NVIDIA UFM Platform, UFM AI Telemetry |
Technical Specifications
| Parameter | Q3200-RA Specification |
|---|---|
| Model Number | Q3200-RA (NVIDIA SKU: 920-9B38F-00RX-6S0) |
| Port Density | 72 NDR 400Gb/s ports (2 independent switch systems x 36 OSFP cages) |
| Form Factor | 2U Rackmount |
| Aggregate System Throughput | 28.8 Tb/s |
| Management Ports | FNM (OSFP), 2x MGT (RJ45), Console (RJ45), USB 2.0 |
| Power Supplies | 6 Hot-swappable, 1500W AC (200-240V) or 48V DC options |
| Cooling | 12 Hot-swappable fan units, Connector-to-Power (C2P) Airflow (Front-to-Rear) |
| Dimensions (HxWxD) | 3.4" x 17.2" x 30.5" (86.4mm x 437mm x 775mm) |
| Weight | Approx. 32 kg (fully loaded with 6 PSUs) |
| Operational Temperature | 0°C to 40°C (depending on optical module usage) |
Specifications are subject to verification. Please confirm exact configuration with your sales representative.
Buyer Checklist
- Rail Kit Required? Yes (Standard 2U sliding rails)
- Airflow Direction: C2P (Connector to Power). Ensure cold aisle is on the port side
- Minimum OS Version: MLNX-OS 3.10.x or later
Service & Support
- 1-Year Standard Hardware Warranty (Extendable to 3 or 5 years)
- NVIDIA Firmware Update Access
- 24/7 Technical Email & Phone Support
- Advance Replacement for DOA within 30 days
- Free Deployment Consultation (for bulk orders)
Company Introduction
As a certified NVIDIA networking partner, we specialize in the global distribution of InfiniBand and Ethernet switching solutions. We focus on AI, HPC, and Cloud data centers, providing authentic hardware with logistics support, pre-sales engineering, and after-sales service tailored to enterprise requirements.
Related Products
- NVIDIA Q3400-RA (4U 144-Port XDR 800Gb/s)
- NVIDIA ConnectX-7 NDR Adapter Cards
- NVIDIA Quantum-2 MQM9700 Series
- OSFP 400Gb/s Passive DAC and AOC Cables
Related Guides
- Designing Cost-Effective Leaf-Spine Fabrics with Q3200-RA
- NVIDIA UFM for AI Cluster Management
- Understanding C2P vs P2C Airflow in Data Centers
Frequently Asked Questions
Q: Is the Q3200-RA two separate switches or one logical switch?
A: It contains two physically independent 36-port 400Gb/s switches. They share the same chassis and power supplies but operate as separate fabric planes.
Q: Can I use the Q3200-RA in a spine role?
A: Yes. It is excellent for spine roles in smaller clusters or for partitioning larger fabrics. For very large spine layers, the Q3400-RA (144 ports) is recommended.
Q: What is the typical power consumption?
A: Approximately 1.5 kW with passive copper cables, up to 4 kW with active optical modules. Exact values depend on PSU configuration.
Q: Does the Q3200-RA support in-service software upgrade?
A: Yes, supported with redundant management modules. Please verify feature set in the specific MLNX-OS version.
Important Notice
This product is intended for professional data center deployment only. Use with NVIDIA-approved rail kits and power cords. Airflow direction (C2P) must match your rack hot/cold aisle configuration. Specifications, warranty terms, and software support may vary by region. Please consult your sales representative for definitive information before ordering.
Vuoi conoscere maggiori dettagli su questo prodotto