NVIDIA 920-9N301-00FB-0C0 | Spectrum-3 32-Port 400GbE Switch, 12.8Tbps

Q: What deployment scenarios is the MSN4700-WS2FC optimized for?
A: This switch is purpose-built for data center spine and super-spine roles in leaf-spine fabrics, AI/ML training clusters with GPU interconnect requirements, hyperscale cloud networks, and high-performance computing environments requiring low-latency east-west traffic and RoCEv2 support for storage over Ethernet.

Q: Can the 400GbE ports be configured for lower speeds or breakouts?
A: Yes, each QSFP-DD port supports flexible speeds from 1GbE to 400GbE and multiple breakout configurations including 4x100GbE, 2x200GbE, and 8x50GbE per port, enabling migration-ready designs and connectivity to endpoints at varying speeds using appropriate breakout cables and transceivers.

Q: What operating systems are compatible with this switch?
A: The MSN4700-WS2FC ships with Cumulus Linux pre-installed and also supports other open network operating systems including SONiC and ONIE-compatible distributions. The switch runs on an x86 architecture with a quad-core 2.20GHz CPU and 16GB RAM for flexible automation and orchestration workflows.

Q: Does this switch support RDMA over Converged Ethernet for storage and HPC workloads?
A: Yes, the Spectrum-3 ASIC provides hardware-accelerated RoCEv2 support with lossless Ethernet capabilities, making it ideal for NVMe over Fabrics, GPU Direct Storage, and distributed AI training applications requiring ultra-low latency and high throughput for east-west traffic.

Q: What network virtualization and telemetry features are included?
A: The switch supports VXLAN routing with hardware acceleration, IPv6 segment routing, cloud-scale NAT, deep packet inspection up to 512 bytes, and NVIDIA What Just Happened (WJH) telemetry for real-time visibility into network events, plus streaming telemetry and in-band network telemetry (INT) for programmable monitoring.