Hello community!
Hopefully someone here can help me with my question. I am basically looking for the configuration required on the ConnectX6-DX NIC to change it from 1x100G Ethernet to 4x25G Ethernet. I have found information on how to do this on the switch here, but I was wondering how to do this on the server side NIC instead (if possible!).
Thanks for your reply @dwaxman . Is there another NIC model from Nvidia/Mellanox that supports this feature? I know for a fact that Emulex/Broadcom have this kind of feature (break-out cable support) in some of their NICs because I used in the past.
We do also actively use such a mode with the Intel cards.
It would have been nice to have another option in case there is a shortage, but ConnectX is ruled out then.
So, MY case is where I’m using SuperMicro “SuperServers”/TwinPro 2U 4node servers as my cabinet router/“switches” and with 2x 100G QSFP28 -breakout to 4x25G SFP28 ports, I suddenly have 2x 2x 4xSFP28 (16xSFP28) possible ports per node to connect the mesh routing to the rest of the servers in the cabinet (typically a half rack, and yes, space/budget constraints)
But that is my usage scenario where the switches would be superfluous extra hops not wanted/needed in the meshed cluster setups