Brand: NVIDIA | Category: Network Switches
SKU: QM9900 | Part #: QM9900 | MPN: QM9900
Contact for Pricing — Request a Quote
The NVIDIA Quantum-3 InfiniBand switch (QM9900) delivers redundant hot-swappable AC power supplies and hot-swappable redundant fans with front-to-rear airflow, ensuring maximum uptime and operational resilience for mission-critical data center environments. This 1U rack-mountable switch represents the latest generation of NVIDIA's high-performance fabric technology, engineered for hyperscale AI, HPC, and enterprise data centers demanding ultra-low latency and exceptional throughput.
Network engineers and AI infrastructure teams will recognize the QM9900 as a cornerstone switching solution that scales to meet the most demanding workloads. With 64 QSFP112 NDR400 ports delivering 400Gb/s bidirectional per port, this switch achieves an aggregate switching bandwidth of 51.2Tb/s full duplex—enabling seamless multi-GPU and multi-node communication. The switch supports NDR400, NDR200 via splitter cable, and HDR100 backward compatibility, making it a future-proof investment. Sub-microsecond port-to-port cut-through latency combined with NVIDIA SHARP v3 in-network computing capabilities accelerates collective communications and distributed applications. Hardware-based adaptive routing with real-time congestion awareness and InfiniBand congestion control with QoS ensure optimal fabric performance under load. The QM9900 integrates seamlessly with NVIDIA ConnectX-7, BlueField-3, and prior-generation InfiniBand host channel adapters, supporting fat-tree, dragonfly, and enhanced hypercube network topologies. Management is simplified through an out-of-band Ethernet management port and full compatibility with NVIDIA Unified Fabric Manager (UFM). The switch complies with the InfiniBand Architecture Specification NDR standard.
| Brand | NVIDIA |
| Category | Network Switches |
| SKU | QM9900 |
| Part Number | QM9900 |
| Condition | New |
| Capacity | 800GB |
| Speed | 800Gb/s |
| Manufacturer Part Number | QM9900 |
| Product Line | Quantum-3 InfiniBand |
| Switch ASIC | NVIDIA Quantum-3 |
| InfiniBand Generation | NDR (Next Data Rate) |
| Number of Ports | 64 x QSFP112 NDR400 ports |
| Per-Port Bandwidth | 400Gb/s bidirectional per port |
| Aggregate Switch Bandwidth | 51.2Tb/s full duplex |
| Port Speed Options | NDR400 (400Gb/s), NDR200 (200Gb/s, splitter cable), HDR100 backward compatible |
| Latency | Sub-microsecond port-to-port cut-through latency |
| In-Network Computing | SHARP v3 (Scalable Hierarchical Aggregation and Reduction Protocol) |
| Adaptive Routing | Hardware-based adaptive routing with real-time congestion awareness |
| Form Factor | 1U rack-mountable |
| Power Supplies | Redundant hot-swappable AC power supplies |
| Cooling | Hot-swappable redundant fans, front-to-rear airflow |
| Management Interface | Out-of-band Ethernet management port; NVIDIA Unified Fabric Manager (UFM) compatible |
| Supported Topologies | Fat-tree, dragonfly, enhanced hypercube |
| Congestion Control | Hardware-based InfiniBand congestion control and QoS |
| Ecosystem Compatibility | NVIDIA ConnectX-7, BlueField-3, and prior-generation InfiniBand HCAs |
| Standards Compliance | InfiniBand Architecture Specification NDR |
The NVIDIA Quantum-3 InfiniBand - 800Gb/s (QM9900) is enterprise networking hardware for data-center, campus, and branch deployments. Common roles include core/distribution switching, server uplinks, top-of-rack fabric, and edge connectivity in mixed Cisco/Arista environments.
Key specifications for the NVIDIA Quantum-3 InfiniBand - 800Gb/s (QM9900): new condition; network type InfiniBand Switch; network port config 64x 800Gb/s NDR; network speed 800Gb/s; network switching capacity 102.4 Tbps; network forwarding rate 76.2 Bpps; network power consumption 1200W. Manufacturer part number QM9900. For the full datasheet with electrical, environmental, and compliance details, contact our pre-sales engineering team.
The NVIDIA Quantum-3 InfiniBand - 800Gb/s (QM9900) interoperates with standard switching protocols (LACP, MLAG/VPC, 802.1Q, OSPF/BGP). For specific cross-vendor scenarios — Cisco-to-Arista MLAG peering, fabric integration with Cumulus/SONiC, or migration from legacy Catalyst — our network engineers map a compatibility plan as part of the RFQ.
