MCP4Y10-N002 800G Twin-Port OSFP Passive DAC Cable, 2m, Finned Top – Quantum-2 Switch-to-Switch
Warranty: Lifetime
Twin-port 800G OSFP passive DAC for NVIDIA Quantum-2 switch-to-switch links. Finned top on both ends.
Part number: MCP4Y10-N002 (NVIDIA / Mellanox LinkX)
Description: 800Gb/s twin-port OSFP to twin-port OSFP passive direct attach copper cable, 2×400G NDR InfiniBand, 2 m (6.6 ft), 28AWG
The MCP4Y10-N002 links two NVIDIA Quantum-2 switch ports at 800Gb/s. Each OSFP end carries two independent 400G NDR links (8 × 100G-PAM4 lanes), so one cable joins two switch cages and carries two 400G ports' worth of traffic. Both ends have the finned-top heat sink that air-cooled Quantum-2 and Spectrum-4 switch cages require.
Where It Fits, and Where It Does Not
- Yes: Quantum-2 QM9700 / QM9790 InfiniBand switch to switch (leaf to spine, inter-switch links) inside a rack or to the next rack.
- Yes: Spectrum-4 800GbE switch to switch, where the twin-port OSFP cages are finned-top as well.
- No: DGX H100 / DGX H200 or ConnectX-7 adapters. Those cages are flat-top and their ports are 400G, so a finned-top twin-port cable will not seat. To reach servers from the switch, use the MCP7Y00-N002 800G → 2× 400G breakout DAC, or flat-top optics such as the MMA4Z00-NS400.
Technical Specifications
| Part number | MCP4Y10-N002 |
| Connectors | Twin-port OSFP, both ends |
| Data rate | 800Gb/s (2 × 400G NDR InfiniBand or 800GbE), 8 × 100G-PAM4 |
| Top style | Finned top, both ends |
| Cable type | Passive direct attach copper |
| Wire gauge | 28AWG |
| Length | 2 m (6.6 ft) |
| Hosts | NVIDIA Quantum-2 QM9700 / QM9790, Spectrum-4 twin-port OSFP switches |
| Warranty | Lifetime, backed by Resilient Tec |
Good to Know
- Need 1 m instead? See the MCP4Y10-N001.
- Beyond 2 m, use a finned-top MMA4Z00-NS at each switch with MPO-12/APC OM4 trunks.
- Passive copper draws almost no power and adds the least latency of any switch-to-switch option.
Related Guides
H100 / H200 interconnect matrix: every link in a Hopper cluster, with the right cable for each.
OSFP flat-top vs finned-top: which do you need?
Twin-port 800G OSFP passive DAC for NVIDIA Quantum-2 switch-to-switch links. Finned top on both ends.
Part number: MCP4Y10-N002 (NVIDIA / Mellanox LinkX)
Description: 800Gb/s twin-port OSFP to twin-port OSFP passive direct attach copper cable, 2×400G NDR InfiniBand, 2 m (6.6 ft), 28AWG
The MCP4Y10-N002 links two NVIDIA Quantum-2 switch ports at 800Gb/s. Each OSFP end carries two independent 400G NDR links (8 × 100G-PAM4 lanes), so one cable joins two switch cages and carries two 400G ports' worth of traffic. Both ends have the finned-top heat sink that air-cooled Quantum-2 and Spectrum-4 switch cages require.
Where It Fits, and Where It Does Not
- Yes: Quantum-2 QM9700 / QM9790 InfiniBand switch to switch (leaf to spine, inter-switch links) inside a rack or to the next rack.
- Yes: Spectrum-4 800GbE switch to switch, where the twin-port OSFP cages are finned-top as well.
- No: DGX H100 / DGX H200 or ConnectX-7 adapters. Those cages are flat-top and their ports are 400G, so a finned-top twin-port cable will not seat. To reach servers from the switch, use the MCP7Y00-N002 800G → 2× 400G breakout DAC, or flat-top optics such as the MMA4Z00-NS400.
Technical Specifications
| Part number | MCP4Y10-N002 |
| Connectors | Twin-port OSFP, both ends |
| Data rate | 800Gb/s (2 × 400G NDR InfiniBand or 800GbE), 8 × 100G-PAM4 |
| Top style | Finned top, both ends |
| Cable type | Passive direct attach copper |
| Wire gauge | 28AWG |
| Length | 2 m (6.6 ft) |
| Hosts | NVIDIA Quantum-2 QM9700 / QM9790, Spectrum-4 twin-port OSFP switches |
| Warranty | Lifetime, backed by Resilient Tec |
Good to Know
- Need 1 m instead? See the MCP4Y10-N001.
- Beyond 2 m, use a finned-top MMA4Z00-NS at each switch with MPO-12/APC OM4 trunks.
- Passive copper draws almost no power and adds the least latency of any switch-to-switch option.
Related Guides
H100 / H200 interconnect matrix: every link in a Hopper cluster, with the right cable for each.
OSFP flat-top vs finned-top: which do you need?