NVIDIA Mellanox ConnectX-7 Dual-Port 200GbE QSFP112 SmartNIC MCX755106AS-HEAT PCIe 5.0 x16 - Refurbished

NVIDIA ConnectX-7 Dual-Port 200GbE Adapter Card — Enterprise-Grade AI Networking

Save significantly on the NVIDIA Mellanox ConnectX-7 (MCX755106AS-HEAT), the industry's most advanced SmartNIC for AI, HPC, and hyperscale cloud workloads. This professionally refurbished dual-port 200GbE adapter delivers 400 Gbps total bandwidth with hardware-accelerated networking, storage, and security — backed by Reef Telecom's industry-leading 10-year warranty.

Key Specifications

• Dual-port QSFP112 — 200GbE / NDR200 InfiniBand per port (400 Gbps aggregate)
• PCIe 5.0 x16 host interface (backward compatible with PCIe 4.0/3.0)
• HHHL (Half-Height, Half-Length) form factor with tall and short brackets
• Hardware RoCEv2 with adaptive routing and congestion control
• NVMe/TCP hardware offload acceleration
• GPUDirect RDMA and GPUDirect Storage support
• In-line TLS 1.3 and IPsec crypto acceleration at line rate
• SR-IOV with enhanced VirtIO offload
• Sub-600ns ultra-low latency
• 30W typical power consumption

Ideal Use Cases

• AI/ML training clusters and GPU-dense servers (DGX H100, HGX)
• High-performance computing (HPC) and scientific workloads
• Hyperscale cloud and enterprise data centers
• NVMe-oF and all-flash storage arrays
• High-frequency trading and low-latency financial infrastructure

Why Buy from Reef Telecom?

• 10-Year Warranty — the longest in the industry on refurbished networking equipment
• Free Shipping on orders over $300
• Professionally tested and certified for production deployment
• Compatible with NVIDIA AI Enterprise, Spectrum-3/Spectrum-4 switches, PyTorch, and TensorFlow

The ConnectX-7 is fully compatible with NVIDIA DGX, HGX, and third-party GPU servers. Whether you're scaling an AI training cluster or deploying NVMe over Fabrics storage, this adapter delivers the performance and reliability your infrastructure demands — at a fraction of the cost of new.