NVIDIA Q3400-RA 144-Port 800Gb/s XDR InfiniBand Spine Switch

제품 상세 정보:

브랜드 이름: Mellanox
모델 번호: Q3400-RA (920-9B36F-00RX-8S0)
문서: q3200-ra-q3400-ra-xdr-800gb...al.pdf

결제 및 배송 조건:

최소 주문 수량: 1PCS
가격: Negotiate
포장 세부 사항: 외부 상자
배달 시간: 목록을 기반으로
지불 조건: 티/티
공급 능력: 프로젝트/배치 별 공급
최고의 가격 접촉

상세 정보

모델 번호.: Q3400-RA 920-9B36F-00RX-8S0 전송 속도: 800GB
포트: 72 기능: LACP, 스택 가능, VLAN 지원
기술: 인피니본 최대 속도: XDR
운송 패키지: 포장 사양: 43.6*433.2*590.6 mm
등록 상표: Mellanox 기원: 중국
강조하다:

NVIDIA InfiniBand spine switch

,

144-port high-speed network switch

,

800Gb/s Mellanox XDR switch

제품 설명

NVIDIA Q3400-RA 144-Port 800Gb/s XDR InfiniBand Spine Switch
NVIDIA Quantum-3 XDR 800Gb/s Spine Optimized
Q3400-RA
144-Port XDR 800Gb/s InfiniBand Switch
Built for the largest AI superclusters. The Q3400-RA scales two-level fat-tree topologies beyond 10,000 GPUs with sub-microsecond latency.
115.2
Tb/s Aggregate
144
XDR 800Gb/s Ports
4U
Air Cooled • 8 PSUs
ASICNVIDIA Quantum-3SerDes200Gb/s PAM4ManagementUFM • FNM PortCoolingC2P • 10 Fans
Product Overview
The NVIDIA Quantum-3 Q3400-RA is the industry's highest-density fixed-configuration InfiniBand switch, purpose-built for AI factories and exascale HPC systems. With 144 ports of 800Gb/s XDR connectivity in a 4U chassis, it eliminates the need for multiple spine tiers in large-scale leaf-spine architectures.
Each port operates at 800Gb/s bidirectionally, enabling a single Q3400-RA to connect up to 144 compute or storage nodes at wire speed. When paired with NVIDIA ConnectX-8 adapters, the switch delivers end-to-end 800Gb/s fabric performance.
Technology Deep Dive
The Q3400-RA is powered by the NVIDIA Quantum-3 ASIC, integrating 200Gb/s SerDes technology. Key innovations include:
  • Adaptive Routing for congestion avoidance
  • SHARPv3 in-network compute acceleration
  • Performance isolation for multi-tenant AI clouds
  • Optional router functionality for multi-site fabrics
Key Features
  • 144 XDR Ports - 72 OSFP cages supporting 800Gb/s per port or 2x 400Gb/s per cage
  • 115.2 Tb/s Throughput - Non-blocking switching capacity for massive parallel workloads
  • SHARPv3 Support - In-network data reduction for collective operations, reducing job completion time
  • Fabric Management Port - Dedicated FNM (OSFP) for NVIDIA UFM telemetry and out-of-band fabric orchestration
  • Redundant Power - 8 hot-swappable 2000W PSUs with N+1 or N+N redundancy
  • Ultra-Low Latency - Port-to-Port latency under 130ns
Compatibility Matrix
CategoryCompatible Products
Host Channel AdaptersNVIDIA ConnectX-8 (800Gb/s), ConnectX-7 (NDR 400Gb/s)
Passive CablesOSFP to OSFP DAC (1-3m), OSFP to 2x OSFP splitter
Active CablesOSFP AOC (up to 100m), MPO-to-OSFP active optical modules
Management SoftwareNVIDIA UFM Platform, UFM AI, MLNX-OS
Technical Specifications
ParameterQ3400-RA Specification
Port Configuration144 XDR 800Gb/s ports (72 OSFP cages)
Switching Capacity115.2 Tb/s aggregate, non-blocking
Form Factor4U rackmount, air-cooled, C2P airflow
Management PortsFNM (OSFP), 2x RJ45 (10/100/1000), Console RJ45, USB 2.0
Power Supplies8x hot-swappable, 2000W AC (200-240V), Platinum efficiency
Cooling Fans10x hot-swappable fan units, front-to-rear (C2P)
Dimensions (HxWxD)7.0" x 17.24" x 33.5" (177.8 x 438 x 850 mm)
Weight (full loaded)Approx. 60 kg (132 lbs)
Operating Temperature0°C to 35°C (derated with optical modules)
Verify all specifications with your sales contact prior to ordering.
Buyer Checklist
  • Rail Kit: Required (NVIDIA P/N: 930-9SKIT-00L0-00B for 700-800mm depth)
  • Airflow: Confirm C2P (Connector-to-Power) matches your hot aisle location
  • Power Cords: Not included -- order region-specific C19-C20 cables separately
  • Licensing: UFM or advanced features may require separate license
Service & Support
  • ✓ 1-year manufacturer warranty (extendable to 3 or 5 years)
  • ✓ Firmware and software update access
  • ✓ 24/7 technical support ticket system
  • ✓ Advance replacement for defective units (with support contract)
  • ✓ On-site deployment engineering (available for volume orders)
Related Products
  • • NVIDIA ConnectX-8 800Gb/s Adapter Cards
  • • NVIDIA Quantum-2 NDR (400Gb/s) Switch Series
  • • OSFP 800Gb/s DAC and Active Optical Cables
  • • NVIDIA UFM Software Platform
Related Technical Guides
Deploying XDR Fabrics for 10k+ GPU Clusters • Understanding NVIDIA UFM for AI Observability • C2P vs P2C Airflow in High-Density Switches • SHARPv3 In-Network Compute Use Cases
Frequently Asked Questions
Q: Is the Q3400-RA backward compatible with NDR (400Gb/s) or HDR (200Gb/s) equipment?
A: Yes. The Quantum-3 ASIC supports multi-speed operation. Ports can operate at 400Gb/s or 200Gb/s when connecting to legacy adapters or switches.
Q: Can I use the Q3400-RA as a leaf switch?
A: While possible, it is primarily optimized as a spine switch due to its high 144-port radix. For leaf roles, consider lower-density Quantum-3 models.
Q: What is the typical power draw at full load?
A: Approximately 2.9 kW with passive copper cables. This can increase to 6-7 kW depending on active optical module usage and temperature conditions.
Q: Does the Q3400-RA support in-service software upgrades?
A: Yes, with redundant management planes. Please check the specific MLNX-OS version for upgrade procedures.
Important Precautions
This product is for professional data center installation only. Use only NVIDIA-approved rail kits and power cords. The C2P airflow (connector-to-power) requires the cold aisle to face the port side. Warranty and support terms vary by region. Specifications may change without notice.
NVIDIA Mellanox certified partner distribution center
Company Introduction
We are a certified NVIDIA partner specializing in InfiniBand and Ethernet networking solutions for AI, HPC, and cloud data centers. Our team provides authentic hardware, global logistics, engineering support, and competitive pricing for enterprise-scale deployments. We serve system integrators, research institutions, and Fortune 500 IT teams worldwide.

이 제품에 대한 자세한 내용을 알고 싶습니다
나는 관심이있다 NVIDIA Q3400-RA 144-Port 800Gb/s XDR InfiniBand Spine Switch 유형, 크기, 수량, 재료 등과 같은 자세한 내용을 보내 주시겠습니까?
감사!
답변 기다 리 겠 습 니 다.