NVIDIA Mellanox ConnectX-7 200GbE Dual-Port OCP3.0 Adapter

The NVIDIA Mellanox ConnectX-7 200GbE Dual-Port OCP 3.0 Adapter (MCX713106AS-VEAT) is built on the ConnectX-7 ASIC, NVIDIA's seventh-generation Smart NIC silicon, delivering up to 200 Gigabit Ethernet per port across two ports in an OCP 3.0 Small Form Factor (SFF) mezzanine footprint. The adapter leverages PCIe 5.0 host connectivity to eliminate upstream bandwidth bottlenecks, and supports the full suite of RDMA over Converged Ethernet (RoCE v2) offloads, enabling near-zero CPU overhead for latency-sensitive storage and inter-node messaging workloads. Hardware-based congestion control, adaptive routing, and end-to-end QoS mechanisms are implemented directly in silicon.

ConnectX-7 introduces hardware-accelerated cryptography (IPsec and TLS inline encryption/decryption), advanced telemetry streaming via the NVIDIA Scalable Hierarchical Aggregation and Reduction Protocol (SHARP) framework, and SR-IOV with up to 4,096 virtual functions per port to support dense multi-tenant virtualisation environments. The OCP 3.0 form factor integrates directly onto the server motherboard mezzanine slot, eliminating a PCIe slot and reducing cable complexity, making it the preferred NIC architecture for blade, sled-based, and hyperscale open-rack server designs.

The MCX713106AS-VEAT is engineered for environments running NVMe-oF, distributed AI/ML training fabrics, high-performance computing (HPC) clusters, and software-defined storage platforms such as Ceph and IBM Spectrum Scale. Its support for NVIDIA GPUDirect RDMA enables direct peer-to-peer memory transfers between GPU framebuffers and remote storage or network peers without host CPU involvement, substantially accelerating data pipeline throughput in GPU-dense compute nodes.

Ideal for

• AI and machine learning training clusters requiring high-bandwidth, low-latency GPU-to-GPU and GPU-to-storage communication via GPUDirect RDMA across OCP-form-factor compute sleds
• NVMe over Fabrics (NVMe-oF) storage fabrics connecting all-flash arrays to compute nodes at 200GbE line rate with RoCE v2 offload reducing host CPU storage I/O overhead
• High-performance computing (HPC) workloads such as finite element analysis, computational fluid dynamics, and molecular dynamics that demand tight MPI latency and high message rates between nodes
• Hyperscale and cloud-native data centre deployments using OCP-compliant rack architectures where mezzanine NIC integration is required to maximise chassis density and reduce cabling
• Multi-tenant virtualised infrastructure requiring SR-IOV virtual function assignment to hundreds of VMs or containers per host with hardware-enforced isolation and QoS
• Encrypted east-west traffic in zero-trust enterprise networks leveraging inline IPsec and TLS hardware offload to maintain full line-rate throughput without software encryption penalties

Technical specifications

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