ConnectX-7 Adapter User Manual
Use the exact adapter manual for port implementation, card configuration and supported platform behavior.
PLATFORM GUIDE · NVIDIA CONNECTX
ConnectX Optics and Cable Compatibility Planning Guide
Start from the exact adapter—not from a generic “400G” or “800G” requirement. ConnectX generations, OPNs and port implementations differ, and the peer switch can use a different connector ecosystem.
Exact AdapterGeneration + OPN + port implementation
ProtocolEthernet or InfiniBand configuration
Peer DeviceSwitch family, cage and port mode
Validated InterconnectNVIDIA-supported cable / transceiver evidence
ConnectX-7 is not one connectorNVIDIA documents OSFP and QSFP112 adapter variants.
Switch and adapter ends can differTwin-port OSFP switch cages are not the same as adapter endpoints.
Rate is not compatibility400G alone does not identify the required shell, cable or optic.
Validate against current evidenceUse NVIDIA release notes and supported interconnect lists.
Selection Workflow
Seven gates before selecting a ConnectX interconnect
The objective is to turn a marketing-speed request into a traceable endpoint-to-endpoint configuration.
01 / ADAPTERIdentify exact ConnectX modelRecord generation, OPN, card type and port count.
02 / PORTConfirm physical interfaceOSFP, QSFP112 or another documented endpoint implementation.
03 / PROTOCOLConfirm Ethernet or InfiniBandUse the intended link type and platform configuration.
04 / PEERIdentify the switch endpointRecord switch family, exact cage, port and target mode.
05 / TOPOLOGYConfirm native or breakout pathDocument straight-through, 1:2 or other supported topology.
06 / MEDIAMatch the documented interconnectDAC, transceiver + fiber, or a specifically supported assembly.
07 / RELEASECheck current validated supportReview firmware release notes and NVIDIA supported cable / module evidence.
Connector Reality
Why “400G ConnectX” is not enough information
ConnectX-7 examples
- NVIDIA documents ConnectX-7 variants with single-port OSFP endpoints.
- NVIDIA also documents ConnectX-7 variants with QSFP112 endpoints.
- Exact port splitting capability depends on the supported OPN and configuration.
- Protocol capability and port configuration must be tied to the actual adapter.
Peer-side differences
- Quantum-2 NDR and Spectrum-4 air-cooled switches use twin-port OSFP cages in NVIDIA's 100G-PAM4 ecosystem.
- Adapter-side OSFP is a single-port endpoint and should not be confused with twin-port switch OSFP implementations.
- QSFP112-based adapter endpoints follow a different mechanical path.
- The cable / transceiver must bridge the exact two endpoint implementations.
Compatibility boundary: QIANWU does not label a generic third-party optic or cable “ConnectX compatible” from rate and connector name alone. Compatibility is tied to the exact adapter, peer device, configuration and supporting evidence.
Official Evidence
Use NVIDIA documentation as the platform authority
Current vendor documentation should take precedence over assumptions inherited from another ConnectX generation or another switch family.
Validated Cables and Modules
Firmware release notes maintain validated and supported cable / module evidence for the adapter release.
LinkX 100G-PAM4 Guide
Use NVIDIA's interconnect guide to understand switch-side versus adapter-side connector ecosystems.
Have a ConnectX interconnect requirement?
Send the exact adapter model / OPN, peer switch, port mode, reach, quantity and target deployment date.