The NVIDIA MCP4Y10-N002 is a high-performance 2x400Gb/s twin-port OSFP to 2x400Gb/s OSFP passive Direct Attached Copper (DAC) cable. It delivers an aggregate bandwidth of up to 800Gb/s per twin-port interface by leveraging 8 high-speed electrical lanes operating with 100G-PAM4 modulation per lane. Designed for scalability and power efficiency, this DAC eliminates active optical components, resulting in minimal power consumption (<0.1W typical) and the lowest possible latency — ideal for bandwidth-intensive HPC, AI clusters, and cloud data centers.

Fully compliant with SFF-8665 and SFF-8636 management interface, the MCP4Y10-N002 supports both InfiniBand NDR and 800G Ethernet protocols, auto-detecting the link partner’s capability. Its 2-meter length with 26AWG conductors offers robust signal integrity while maintaining excellent bend flexibility for dense cabling environments.

NVIDIA’s passive DAC architecture eliminates retimers and optical transceivers, providing a direct copper connection that minimizes serialization latency — a critical advantage for AI training and high-frequency trading workloads. Each cable assembly is length-tuned to reduce internal reflections and crosstalk, ensuring robust eye opening at 100G-PAM4 per lane. The MCP4Y10 series uses precision impedance control (100Ω differential) and superior shielding to meet stringent 800G signal integrity requirements. Built-in EEPROM memory maps store vital product data such as vendor information, part number, serial number, and cable length, enabling seamless system management via the host.

Technology highlight Twin-port OSFP design doubles port density while maintaining backward compatibility with existing OSFP cages, facilitating higher bandwidth per rack unit.

• Switch-to-Switch interconnection: Connect two Quantum-2 InfiniBand switches or Spectrum-4 Ethernet switches at 800Gb/s with low cost and latency.
• Switch-to-DGX H100 / HGX H100 systems: Provide high-throughput fabric links in NVIDIA DGX H100 clusters using flat-top connector variants (FLT) for liquid-cooled racks.
• Spine-leaf architectures: Deploy in high-performance data centers for scalable, power-efficient leaf-spine topology.
• AI & HPC clusters: Ideal for large-scale AI training where ultra-low latency and near-zero power per link are critical.

The MCP4Y10-N002 is natively compatible with NVIDIA Quantum-2 InfiniBand switches (e.g., QM9700 series) and NVIDIA Spectrum-4 Ethernet switches (SN5600/SN5700 series). It supports both InfiniBand NDR and 800G Ethernet modes without firmware changes — the cable auto-configures based on the connected switch ASIC. It also works with third-party 800G OSFP ports that adhere to OSFP MSA and 100G-PAM4 electrical interfaces (please confirm interoperability with non-NVIDIA systems).

Note: Specifications are based on published datasheet revision 1.2 (Sep 2023). For custom lengths or flat-top option, refer to ordering guide.

For DGX H100 to switch connections requiring flat-top on one end, please contact sales for custom assembly. All cables are rigorously production tested for out-of-box reliability.

• Cost-efficiency: Passive DAC eliminates expensive optical transceivers and retimers.
• Negligible power consumption: Nearly zero power (<0.1W) reduces data center cooling and OPEX.
• Lowest latency: Direct copper path avoids serialization/deserialization delays.
• High reliability: Simple passive design with fewer failure points; EMI performance with robust shielding.
• Flexible cable management: Thin 26AWG construction (8.9mm diameter) with excellent bending radius simplifies routing in high-density racks.

As an authorized partner of high-speed networking solutions, we provide full technical support including compatibility validation, deployment guidance, and post-sales warranty services. Each cable includes a standard limited warranty. For large-scale deployments, our team offers pre-sales consultation, cable length customization, and RMA support. Reach out to our technical sales engineers for Bill of Materials (BOM) optimization and site survey assistance.

ESD Protection: The cable is shipped with protective caps. Always follow standard ESD handling procedures—use grounded wrist straps and ESD-safe flooring when inserting or extracting OSFP connectors. Bend Radius: Avoid sharp bends near the connector backshell; maintain at least 10x cable diameter. Installation Guidelines: Refer to the NVIDIA Cable Management Guidelines and FAQ Application Note to prevent physical damage and ensure optimal performance. Do not pull or twist the cable beyond recommended tension.

Regulatory Notices: This is a Class A product. In residential environment it may cause radio interference. Installation must comply with local regulations.

• Confirm required cable length: 0.5m, 1m, 1.5m, or 2m.
• Check connector type: finned-top (air-cooled) or flat-top (liquid-cooled/DGX).
• Verify switch compatibility (Quantum-2 / Spectrum-4 / other vendor).
• Evaluate bend radius and rack depth: minimum assembly space L ~156mm for 2m.
• Ensure ESD-safe installation environment.

• NVIDIA MCP4Y10-N00A – 0.5m twin-port OSFP DAC
• NVIDIA MCP4Y10-N001 – 1m twin-port OSFP DAC
• NVIDIA MCP4Y10-N01A – 1.5m twin-port OSFP DAC
• NVIDIA MCP4Y10-N00A-FLT – 0.5m flat-top OSFP for DGX H100
• NVIDIA MMS4V00-WM – 800G OSFP SR8 optical transceivers for longer reach

• NVIDIA Cable Management Guidelines and FAQ Application Note
• OSFP MSA Specification Rev 1.12
• InfiniBand NDR Architecture Overview
• Deploying 800G Ethernet in Data Centers – Best Practices

For technical documentation or certifications (CB, CE, FCC reports), please contact our support team or request via FAE.