NVIDIA Spectrum SN3700 32-port 200GbE Switch

The Mellanox Spectrum SN3700 is a 1U open Ethernet switch delivering 32 ports of 200GbE (QSFP56) in a non-blocking architecture built on the Mellanox Spectrum-2 ASIC. With an aggregate switching capacity of 12.8 Tbps and a forwarding rate of 9.52 Bpps, the SN3700 is engineered for the most demanding storage fabrics and high-bandwidth enterprise aggregation tiers where legacy 25/100GbE infrastructure has become a throughput bottleneck. The platform supports breakout configurations enabling deployment as a 64-port 100GbE or 128-port 50GbE switch, providing flexible migration paths without forklift upgrades.

The Spectrum-2 ASIC underpins advanced telemetry and congestion management capabilities, including Mellanox's What Just Happened (WJH) streaming telemetry, In-band Network Telemetry (INT), and hardware-based Advanced Congestion Control (ACC). These features provide real-time per-flow visibility and lossless fabric behavior essential for NVMe-oF storage networks, RDMA over Converged Ethernet (RoCE), and GPU-to-GPU communication in AI/ML clusters. The switch supports 4,096 VLANs and 128,000 IPv4 unicast routes, accommodating large enterprise routing tables at the aggregation layer.

The MSN3700-VS2F variant ships with the Cumulus Linux network operating system, enabling a Linux-native, DevOps-friendly operational model that integrates with standard automation frameworks including Ansible, Puppet, and Chef. The platform also supports NVIDIA Air simulation for offline network validation. Dual redundant hot-swap power supplies and hot-swap fans with front-to-rear airflow (VS2F suffix denotes standard airflow) ensure the operational resilience required in production data center deployments.

Ideal for

• NVMe-oF and all-flash storage fabric interconnect requiring lossless RoCEv2 transport with hardware-enforced Priority Flow Control (PFC) and Explicit Congestion Notification (ECN)
• High-performance computing and AI/ML GPU cluster inter-node networking where 200GbE per server port eliminates fat-tree bandwidth bottlenecks
• Enterprise data center aggregation layer replacing stacked 10/25GbE top-of-rack tiers with a single 200GbE aggregation tier supporting large L3 routing tables
• Software-defined networking environments leveraging Cumulus Linux for BGP-based EVPN/VXLAN overlay fabrics with automation pipeline integration
• Financial services low-latency trading infrastructure requiring predictable sub-microsecond cut-through switching and granular per-flow telemetry
• Hyperconverged infrastructure (HCI) scale-out clusters where node-to-node storage replication and compute traffic compete for shared bandwidth

Technical specifications

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