NVIDIA Mellanox 400Gb/s NDR OSFP to 2x200G Twin-port DAC 1.5m

The NVIDIA Mellanox MCP7Y00-H001 is a 400Gb/s NDR OSFP to 2×200Gb/s OSFP passive copper Direct Attach Cable (DAC) measuring 1.5 metres in length. Designed as a splitter breakout cable, it presents a single 400Gb/s NDR OSFP interface on one end and bifurcates the signal into two independent 200Gb/s NDR200 OSFP connectors on the other, enabling a single NDR switch port to serve two GPU nodes or two network endpoints simultaneously. The cable operates passively without active signal conditioning, relying on the SerDes within the connected host adapters and switches to maintain signal integrity at these line rates.

The MCP7Y00-H001 is engineered specifically for high-density AI and HPC fabric deployments where InfiniBand NDR infrastructure must interface directly with NDR200 GPU servers such as those based on NVIDIA ConnectX-7 or BlueField-3 adapters running in 200Gb/s mode. The passive copper construction delivers ultra-low latency and negligible power consumption compared to active optical alternatives, making it well suited for within-rack and adjacent-rack connectivity in tightly coupled GPU clusters. It is compliant with the InfiniBand Trade Association NDR specification and integrates natively with NVIDIA Quantum-2 NDR switches.

At 1.5 m, the cable is optimised for top-of-rack switch-to-server connections inside standard 600–800 mm deep data centre cabinets. Its OSFP form factor on all three connector ends ensures compatibility with the high-power, high-density port designs found on contemporary NDR switching and adapter hardware. The cable is manufactured and tested by NVIDIA Mellanox to operate with NVIDIA networking ecosystem components, enabling plug-and-play recognition and diagnostics through standard cable management tooling.

Ideal for

• Connecting a single NVIDIA Quantum-2 NDR 400Gb/s switch port to two GPU compute nodes each running NDR200 200Gb/s InfiniBand host channel adapters within the same rack
• Building non-blocking fat-tree or dragonfly InfiniBand fabrics for large-scale distributed AI training clusters where GPU-to-switch cable density is a primary design constraint
• Replacing active optical cables in short-reach within-rack segments to reduce power consumption and eliminate transceiver failure points in GPU-dense AI supercomputing pods
• Enabling phased NDR infrastructure deployments where existing NDR200 server nodes are connected to new NDR switching hardware without requiring server-side adapter replacement
• Supporting high-throughput MPI workloads in HPC environments requiring sub-microsecond latency between tightly coupled compute nodes in the same physical rack or adjacent racks
• Providing a passive, low-maintenance interconnect in liquid-cooled high-density GPU racks where thermal budget constraints favour copper DAC over active optical modules

Technical specifications

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