NVIDIA Spectrum SN4700 32-port 400GbE Switch

The NVIDIA Mellanox Spectrum SN4700 is a high-density, non-blocking 400GbE Ethernet switch purpose-built for modern data-center leaf and spine deployments. Based on the NVIDIA Spectrum-3 ASIC, the MSN4700-WS2F delivers 12.8 Tbps of aggregate switching capacity across 32 QSFP-DD ports, each capable of operating at 400GbE, with full support for breakout configurations down to 25GbE and 100GbE per lane. The platform is designed around a deep, shared packet buffer architecture that minimizes congestion-driven packet loss under bursty, mixed workload traffic patterns common in large-scale enterprise and hyperscale environments.

The SN4700 runs Cumulus Linux as its native network operating system, providing a fully open, standards-based Linux environment that integrates with existing automation frameworks including Ansible, Puppet, Chef, and Terraform. The switch supports a rich feature set encompassing BGP Unnumbered, EVPN-VXLAN for multi-tenant overlay networking, RDMA over Converged Ethernet (RoCEv2) with hardware-enforced Quality of Service, ECN, and Priority Flow Control, making it well-suited for latency-sensitive AI/ML training clusters and high-performance computing fabrics. Telemetry capabilities include streaming gNMI/gRPC and SNMP, enabling integration with third-party monitoring and observability platforms.

Physically, the MSN4700-WS2F ships in a 1U form factor with a front-to-rear airflow configuration and redundant, hot-swappable power supplies alongside hot-swappable fan modules, supporting continuous operation without scheduled maintenance windows. The switch is compatible with NVIDIA Air simulation environments and the broader NVIDIA Spectrum ecosystem, allowing operators to validate configurations offline before production rollout. Its combination of port density, buffer depth, and open software makes it a primary candidate for enterprises accelerating data-center fabric modernization toward 400GbE infrastructure.

Ideal for

• High-performance AI and ML training cluster interconnects requiring lossless RoCEv2 fabric with hardware QoS enforcement across GPU server racks
• Enterprise data-center leaf-spine fabric upgrades replacing legacy 100GbE top-of-rack switches with 400GbE-capable high-density alternatives
• Multi-tenant private cloud environments leveraging EVPN-VXLAN overlays for network virtualization and workload isolation across shared physical infrastructure
• High-frequency financial services workloads demanding consistent low-latency switching and deep buffer protection against microbursts
• Large-scale storage networking fabrics connecting NVMe-oF and all-flash arrays where throughput, low latency, and congestion management are critical
• Network automation and NetDevOps environments where open Linux-based operating systems and API-driven management enable CI/CD pipeline integration for network configuration

Technical specifications

Available from Omnixon Global. Submit an RFQ and our team will confirm configuration and availability for your order.