I'm looking at a switch like this:
https://www.fs.com/products/29123.html
It has 48 10-GbE SFP+ ports and 6 40-GbE uplink ports. I would like to use some of these uplinks to cross-connect to another switch. In addition to that, however, I have a server with a 40-GbE capable NIC in it. Can this be connected directly to the 40-GbE port on the switch and "just work," or are these uplink ports special in that they must connect to uplink ports on another switch?
I know this used to be a thing for copper-based cabling before Auto-MDIX became commonplace, but it's not clear to me whether it's still an issue for modern SFP/QSFP-based connections.
Best Answer
That switch uses (Q)SFP+ ports which are essentially PHYless. You need to either use matching PHY type transceivers on both the switch and the host ports (e.g. 10GBASE-SR or 40GBASE-SR4) or direct-attach cables that are compatible with both sides.
The Mellanox NIC has QSFP28 ports that should support QSFP+ modules as well - check the documentation.
Note that most switches or NICs only support correctly branded transceivers - directly from the vendor or 3rd party compatible ones.
Whether the switch connects to a host or to another switch generally doesn't matter, assuming the port functions are configured appropriately.
MDI/MDI-X/Auto MDI-X are a special function of twisted-pair ports. The reason is that TP cables are straight-through historically and the receiver/transmitter crossover happens inside one of the link ports.
(Q)SFP(+) ports are all the same, the crossover happens within the cable (fiber or DAC).