So much misinformation on 10G.
You can still do 2.5G on most 10G NIC. There is little to no price difference on a 10G card today, 10G is “old” so the server equipment is being sold at bottom barrel prices. It’s about $30 for a decent Intel 10G NIC, even a dual-port for $50, the TP-Link 2.5G I can find is also $30, but note the brand, note that TP-Link last time I tried didn’t work with Cisco 2.5/5/10G ports, just a difference in spec interpretation, so it went to 1G.
Running 10G over CAT5e is possible for most ‘home’ environments, the crosstalk protection isn’t there but the frequency rating of 250MHz is so if you have a decent quality cable, especially FTP/STP it should work the same as CAT6 over short distances (<15ft) just make sure your cable is properly crimped.
For long running (>55ft) proper 10G over copper you need CAT6a, unless you are a professional, you don’t have that equipment at home to run CAT6a over those distances. Sorry, it won’t pass the spec test, the spec test machine from Fluke is $15k. Even most pre-molded cables only support CAT6a up to 55m (they are really CAT6). I have trouble all the time with professional cable installers that know CAT6a and still weird issues happen (connection drops every 10 minutes) then we test and they need to redo specific patches, as such, most datacenters now go with fiber, two reasons, easy to support higher speeds when you eventually get there, easier and cheaper to install. The individual conductor tolerance is I believe <2mm over 100m, that means from production of the cable to you installing (bending/pulling) and cutting the cable to crimp it you cannot pull any pair out of alignment or cut it slightly at an angle. There is specs for how hard you can pull on the cable, how much it can bend and how many can be in a conduit is also severely limited because the cable is so thick.
Fiber (multimode) is easy to install, the kit is relatively cheap, once you do a few, you’d be surprised how easy it is. It is easier than CAT5, it’s very similar in process to coax cable. The most fiddly thing about it is the tiny clips to make it into a duplex (it’s just plastic clips, nothing special). Any quality cable will be marked, so for a duplex, just calculate the length and cut it at the marking. Modern multimode is more tolerant to torque and bend than CAT6a, and much cheaper ($250/1000ft (500ft duplex)) than proper CAT6a copper ($300-500/250ft), the NIC and optics are dirt cheap and run less hot, a dual port 520 can be had on eBay for $15, the optics for $8/port in bulk.
That being said, yes, CAT6a runs slightly hotter, unless you are packing 100 in a raceway, you don’t have to worry about that, if your cable gets noticeably warm to the point of melting in your homelab, you have other issues.
MTU9000 (Jumbo Frames) is always weird, these days with interrupt coalescing and decent (server) NIC, you can run 100G or 400G at standard Ethernet frames. It may be slightly more efficient if you consistently need large packets, but it also has trade-offs. All switches and NIC can run full speed and have the pps capacity to run at regular 1500MTU because most of the Internet runs at or below 1500MTU, it’s been since the Pentium 4 era that 10G was still a consideration for jumbo frames.
