NVIDIA 920-9B210-00RN-0M2 | Quantum-2 64-Port NDR InfiniBand Switch, 51.2Tbps

Q: What is NDR InfiniBand and how does it differ from previous generations?
A: NDR (Next Data Rate) InfiniBand operates at 400Gb/s per port, doubling the bandwidth of the previous HDR generation. The MQM9700-NS2R is backward compatible with HDR and earlier InfiniBand standards, allowing seamless integration into existing infrastructures while providing a clear upgrade path.

Q: Can the OSFP ports be split into multiple lower-speed connections?
A: Yes, each of the 32 OSFP ports supports breakout configurations. The switch can be configured to deliver up to 128 ports at 200Gb/s using appropriate breakout cables and port splitting configuration, providing flexibility for different network topologies and connectivity requirements.

Q: What is the difference between C2P and P2C airflow models?
A: The MQM9700-NS2R features Connector-to-Power (C2P) reverse airflow, drawing air from the front connector side and exhausting through the power supply side. This is ideal for hot-aisle/cold-aisle data center designs with reverse airflow requirements. The P2C models provide standard front-to-back airflow.

Q: What network topologies does this switch support?
A: The QM9700 supports a wide variety of InfiniBand topologies including Fat Tree, SlimFly, DragonFly+, and multi-dimensional Torus configurations. The managed variant includes an on-board subnet manager that simplifies deployment for fabrics up to 2,000 nodes without requiring external management infrastructure.

Q: Is this switch suitable for AI and machine learning workloads?
A: Absolutely. The MQM9700-NS2R is purpose-built for AI training clusters, supporting NVIDIA SHARP v3 (Scalable Hierarchical Aggregation and Reduction Protocol) for in-network computing acceleration, RDMA for ultra-low latency data transfers, and adaptive routing to optimize traffic flow for distributed AI workloads such as those running on NVIDIA DGX systems.