NVIDIA 400G OSFP to 2x200G QSFP56 InfiniBand Splitter DAC

The NVIDIA MCP7Y00-N002 is a 400G OSFP to 2×200G QSFP56 InfiniBand NDR splitter Direct Attach Copper (DAC) cable, engineered to break out a single 400G NDR OSFP port into two independent 200G NDR QSFP56 connections. This passive copper architecture delivers ultra-low latency and deterministic signal integrity without requiring active optical components or external power beyond what the host ports supply. At 2 metres in length, it is optimised for top-of-rack and intra-rack interconnects in high-density GPU and accelerated computing environments.

Designed specifically for NVIDIA Quantum-2 NDR InfiniBand infrastructure, the MCP7Y00-N002 enables network architects to maximise port utilisation on NDR 400G switches such as the NVIDIA QM9700 series. By splitting one 400G OSFP switch port into two 200G QSFP56 endpoints, operators can double the number of connected nodes per switch port, a critical capability when building large-scale GPU clusters where every link counts toward all-reduce and collective communication bandwidth. The cable is compliant with InfiniBand NDR signalling standards and is fully supported under NVIDIA's ConnectX-7 and BlueField-3 adapter ecosystem.

As a passive DAC assembly, the MCP7Y00-N002 requires no firmware, no additional drivers, and no thermal management beyond standard rack airflow. Its copper construction ensures deterministic, jitter-free performance suitable for MPI workloads, NCCL-based GPU collective operations, and storage-over-fabric deployments. The splitter form factor makes it a foundational building block for fat-tree, dragonfly, and other high-radix InfiniBand topologies where NDR400 spine switches feed NDR200 leaf or endpoint tiers.

Ideal for

• Connecting dual NDR200 GPU compute nodes to a single NDR400 spine switch port in large-scale AI training clusters, doubling node density per switch port
• Building cost-efficient fat-tree InfiniBand fabrics where NDR400 core switches fan out to NDR200 ConnectX-7 HCA-equipped servers at the edge tier
• High-performance computing (HPC) installations requiring deterministic low-latency MPI communication between GPU nodes within the same rack or adjacent racks
• Intra-rack GPU-to-GPU direct interconnect in hyperscale AI infrastructure where minimising cable plant complexity and eliminating active optical components reduces failure domains
• Storage-over-fabric architectures leveraging NVMe-oF over InfiniBand NDR, where a single high-radix switch port fans out to two NVMe storage server HBAs
• Staged NDR fabric expansions where existing NDR400 switch ports are progressively split to NDR200 endpoints as additional server nodes are phased into a growing cluster

Technical specifications

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