NVIDIA ConnectX-7 200GbE Dual-Port QSFP112 PCIe Gen5 Adapter

The NVIDIA ConnectX-7 200GbE Dual-Port QSFP112 PCIe Gen5 Adapter (MPN: 900-9X718-0076-DT0) is a high-performance network interface card built on NVIDIA's ConnectX-7 silicon, delivering up to 200 Gigabits per second per port across two QSFP112 interfaces for an aggregate throughput of 400Gb/s. The adapter leverages PCIe 5.0 x16 host connectivity to eliminate I/O bottlenecks between the NIC and the host CPU, making it well suited for data-intensive compute and storage environments that demand maximum bandwidth with minimal latency. Hardware-offload engines for RDMA over Converged Ethernet (RoCE), GPUDirect RDMA, and NVMe over Fabrics (NVMe-oF) are integrated directly into the silicon, reducing CPU utilization and enabling line-rate data movement between storage, compute, and GPU resources.

ConnectX-7 introduces support for NVIDIA's NVIDIA-Quantum-2 and BlueField ecosystem, and the adapter includes advanced features such as hardware-based transport offload, adaptive routing support, and in-network computing primitives when deployed within compatible NVIDIA Spectrum or InfiniBand fabrics. The card supports single-root I/O virtualization (SR-IOV) with up to 4,096 virtual functions per port, enabling dense multi-tenant virtualized and containerized deployments. Comprehensive support for VXLAN, NVGRE, GENEVE, and MPLSoUDP/GRE encapsulation offloads ensures compatibility with modern software-defined networking overlays without sacrificing throughput or latency.

Targeted at high-performance computing (HPC) clusters, AI/ML training infrastructure, all-flash NVMe storage arrays, and cloud-scale data center fabrics, the ConnectX-7 200GbE adapter is qualified for deployment in leading server platforms and is supported under major Linux distributions, VMware vSphere, and Windows Server environments. The adapter's support for Open MPI, UCX, and DPDK accelerates both traditional HPC message-passing workloads and modern data-plane applications, providing a future-proof foundation for next-generation 200G and 400G network migrations.

Ideal for

• AI and deep learning training clusters requiring high-bandwidth, low-latency GPU-to-GPU communication via GPUDirect RDMA across a 200GbE RoCE fabric
• High-performance NVMe-oF storage fabrics connecting all-flash arrays to compute nodes at line rate with hardware-offloaded initiator and target operations
• High-performance computing (HPC) environments running MPI workloads that benefit from hardware transport offload and RDMA to minimize inter-node latency
• Dense virtualized or containerized cloud infrastructure leveraging SR-IOV virtual functions and VXLAN/GENEVE offload for scalable multi-tenant network segmentation
• Hyperconverged infrastructure nodes where a single high-throughput adapter consolidates storage, live migration, and VM data traffic onto a unified 200GbE fabric
• Data analytics and in-memory database platforms that require sustained 200Gb/s throughput per port to prevent network I/O from becoming the performance bottleneck

Technical specifications

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