The NVIDIA Mellanox SN3700 delivers next-generation 200 Gigabit switching for spine and super-spine deployments. This refurbished Spectrum-2 switch packs 32 QSFP56 ports into a 1RU form factor, delivering a massive 12.8 Tbps of non-blocking throughput for AI infrastructure, machine learning clusters, and high-density data center environments.

Key Benefits:

• 12.8 Tbps switching capacity handles AI and ML workloads
• 8.33 Bpps forwarding rate for ultra-low latency applications
• 425ns latency keeps distributed training synchronized
• Zero packet loss with 42 MB shared buffer architecture
• Native RoCEv2 support for RDMA storage acceleration
• Flexible 10G to 200G operation on all ports
• 512K forwarding entries for massive scale deployments
• What Just Happened telemetry for rapid troubleshooting

Ideal Use Cases:

• AI training clusters requiring GPU-to-GPU connectivity
• Machine learning infrastructure with distributed workloads
• NVMe over Fabrics storage networks
• Spine and super-spine in leaf-spine architectures
• High-frequency trading and financial applications
• Hyperscale data center core switching
• GPU server top-of-rack with 200G uplinks
• Large-scale VXLAN overlay networks

Technical Specifications:

• 32x QSFP56 ports supporting 10/25/40/50/100/200GbE per port
• 12.8 Tbps total switching capacity
• 8.33 Bpps forwarding rate
• 425ns port-to-port latency
• 42 MB shared packet buffer with predictive congestion control
• Layer 3 routing with BGP, OSPF, VXLAN, and segment routing
• RoCEv2 and GPUDirect for RDMA acceleration
• 512K shared forwarding entries (ACL, routes, MAC, ECMP)
• 10X better VXLAN scale vs. previous generation
• Hardware-accelerated telemetry and histograms
• Dual redundant hot-swappable AC power supplies
• Integrated x86 processor for management
• Standard 1RU depth: 17.32 inches wide x 20.98 inches deep
• Runs NVIDIA Onyx, Cumulus Linux, or ONIE

This refurbished unit has been professionally tested and certified to like-new condition. Includes dual redundant AC power supplies and supports both port-to-cooling (P2C) and cooling-to-port (C2P) airflow configurations for flexible data center thermal management.