I just bought 2 Brocade 6450 off ebay after someone on RedFlagDeals sent me link.
No clue what I've gotten myself into, but I guess I need this to use the 10G ports?
Your common 10G Xcvr will be SFP+, the physical size is still the same as common 1G SFP they started out larger IE CFP but as time marches on the electronics shrink. These 40G Xcvrs have 4 lasers packaged in a tight footprint that is slightly larger than SFP/SFP+ these are physically Quad SFP (QSFP) and you see the newer 100G gear using QSFP28 to indicate the same size but faster signaling rate, there are also SFP28 so the smaller single laser size but able to do 25GE.
When you're dealing with Quad Xcvrs that are indicated as LR4 those will generally support breaking down into 4 separate signals using no active device just a physical cable that separates the 4 carriers, so with these you could have one connected to four 10G endpoints given the correct fibre/DAC pigtail. Similarly QSFP28 can be 1x100G or bifurcated into 4x25G.
Buying surplus can be a pain in posterior though due to the differences in who the specs were designed for and when. In carrier where I work 10G briefly used CFP but leaned heavily into XFP form factors, we use loads of them still with SFP+ coming in at a trickle still. Most of our 100G are still CFP though the new stuff coming in is all QSFP28. We didn't use a whole bunch of 40G and it didn't much matter because our gear breaks those out to 10Gs and when those were hot customers didn't have clients that could handle a full 40G.
More importantly we exclusively use single mode fibre but as data center customers have moved up the speed ladder there have been some annoying sidebars. Short reach quad form factors require multimode MPO bundle fibre for example, which are a pain and time sink to scope/clean and we hate anything where a single fibre failure may require replacement of a larger bundle cable and cause down time for adjacent customers. Thankfully those have also been short lived because DCs charge per fibre pair when placing cross connects so colo customers have pushed through to SM LR4 to save XC MRC charges. I'm getting a little far afield but a relevant example that brought this to mind is 40G. Some 40G ports support 50G rates as well but while you can break a 40G LR4 into 4x 10G LR, if you have a 40/50G card you can't connect that to half a 100G LR4 because the 100G LR4 is pumping out 4x25G channels, so the 40/50G card can't keep up with the signaling rate of two carriers pushing the data rate they can handle with 4 carriers. You could break that 100G into 4x 25G, 2x50G etc. Completely speculative since I don't have 1st hand experience but seem to recall mention of extreme home labbers fiddling with 100G broken and slowed into 4x10G carriers but I wouldn't bet on being able to do that and unless you got the gear on a lark it's a waste of the 100G port price premium. However if the xcvr modules fits it may work in a socket rated faster, so you could potentially use these 40G QSFPs in a QSFP28 socket on a 100G switch if say you wanted to burn one port to connect down stack to 10G/40G ports on smaller switches to bring slower infrastructure into the mix with the 100G monsters.