NVIDIA/Mellanox 200G QSFP56 to 2x100G Split DAC Cable

The Mellanox MCP7H50-V002R26 is a passive Direct Attach Copper (DAC) breakout cable designed to deliver 200 Gigabit Ethernet or HDR100 InfiniBand connectivity from a single QSFP56 port, splitting into two independent 100G QSFP28 connections. Built on a copper twinaxial construction, the cable supports PAM4 modulation on the 200G trunk side and NRZ signaling on the 100G breakout legs, enabling high-density, low-latency interconnects without the power overhead of active optical solutions. The 2-meter length and passive design make it well-suited for top-of-rack and end-of-row cabling scenarios common in high-performance computing and AI/ML infrastructure.

The cable is fully compliant with the QSFP56 MSA specification and integrates with Mellanox Quantum and ConnectX adapter families, including HDR100 InfiniBand switch ports operating at 2×100Gbps per port. Its copper passive architecture introduces no signal conversion latency, making it a preferred choice wherever deterministic, sub-microsecond interconnect latency is critical. The QSFP56 housing on the trunk end carries 8 lanes at 25Gbps each (aggregate 200Gbps), which are electrically bifurcated to two QSFP28 connectors each carrying 4 lanes at 25Gbps (100Gbps per leg).

First introduced in 2020 Q2, the MCP7H50-V002R26 has become a standard bill-of-materials component in GPU cluster fabrics, large-scale machine learning training environments, and financial services HPC racks where rack density and cable management efficiency are paramount. Procurement planning should account for industry-wide extended lead times commonly associated with this product class, particularly during periods of elevated AI infrastructure build-out demand.

Ideal for

• Connecting a 200G HDR100 InfiniBand spine or leaf switch port to two separate 100G server HCAs in GPU compute clusters for AI and deep learning training workloads
• High-density top-of-rack cabling in HPC environments where passive DAC eliminates per-cable power draw and optical transceiver heat management concerns
• Bifurcating 200G switch uplinks to dual 100G storage nodes in NVMe-over-Fabrics (NVMe-oF) all-flash storage architectures
• Financial services low-latency trading infrastructure requiring consistent, hardware-deterministic sub-microsecond switch-to-server interconnect latency
• Disaggregated AI inference rack deployments where a single switch port must serve two independent accelerator trays at full 100G bandwidth each
• Data center migration and fabric consolidation projects transitioning from 100G leaf-spine topologies to 200G spine switches while reusing existing 100G server NICs

Technical specifications

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