Does 25GBase-T Exist?

Networking
1

25GBase-T and 40GBase-T were standardized in 2016 as IEEE 802.3bq to enable 25/40GBit Ethernet over Category 8 cables. Yet I can’t find any evidence that it has ever been implemented in a commercial product. I’ve only ever found mislabeled SFP28 ports or fiber optic transceivers, nothing actually RJ45. I expected that by now there would at least be a few products available even if they were ludicrously expensive.

I’m particularly surprised that 25GBase-T transceivers don’t seem to exist because advertising for Category 8 cables implies that they do. Often the advertising is careful to just say the cables support 25/40Gbit speeds, which isn’t untrue I suppose, but sometimes advertising says that you can use Category 8 cables for 25Gbit Ethernet which appears to be untrue unless you work in a lab at Broadcom. For example, Tripp-Lite:

Cat 8 Ethernet cable is ideal for switch to switch communications in data centers and server rooms, where 25GBase‑T and 40GBase‑T networks are common.

I think it’s fair to say that’s an incorrect statement. I don’t think it’s a deliberate lie to sell Cat-8 cables, I just think the writer doesn’t know that “25GBase‑T” is specifically twisted pair and not a generic term for 25Gbit Ethernet.

Can anyone find a commercial product which supports 25GBase-T or 40GBase-T?

Note: I am not asking if 25GBase-T is a good idea, it’s obviously not in the vast majority of situations. Just use DAC or fiber.

2

Thats rather interesting question, but who would willingly use it?

Industry has leapfrogged even 40GBASE since 2016 and even 10GBAse is extremely inefficient over copper.

A am trying to think about any valid use case vs. going fiber and getting bupkis.

I.e:

  • do you need 1GBE+ speeds? Copper wins here by default.
  • do you need 10GBE+ speeds? Fiber wins here by opex, and capex costs might be same or lower in the end
  • Do you need 25GBE+? If so, fiber is both opex and capex more efficient.

EDIT: I just cannot see any usecase where 25/40GBASE-T isnt demolished by simplicity and cost of sfp DAC.
Other than we need few 10GB+ uplinks and for some reason we cannot afford cheap sfp28/qsfp fiber modules.But FS.com sells cisco compatibe 25GBE fiber modules starting at 40 USD, same for 40GBE , lol :slight_smile: 100GBE you ask? 100 USD for 100m ones.

2 Likes
3

What greatnull said. There is pretty much no reason to go copper for speeds over 10GbE. You need to rewire with CAT8 in any case so why not just do fiber and support up to Tbit speeds*?

Do yourself a favor, invest in fiber if you have a use case for 10 GbE+ :slightly_smiling_face:

*complete with the plaid effect when going at ludicrous speeds ofc :grin:

4

Hyperscalers aren’t using anything under 100Gbps (think mellanox cx6 and mellanox cx7 and their broadcom equivalents). Some are pushing faster than 400Gbps to a card because of ML stuff in various weird topologies.

I don’t see smaller DCs going with copper 25 either, pass me down 25/100G network gear is relatively easy to get, and there’s newer cheaper 25/100 gear showing for that market.

Maybe in the home market in 10 years or so.

2 Likes
5

Also going with copper over 25GBE in enteprise setting is moot point, since there seem no be no qsfp modules available. If they are, they will be proportinally more expensive versus firer ones, due to economy of scale.

Even SFP+ to 10GBASE-T modules are somwhat rare. Have you seen intel DA cards support lists? Fiber modules only.

Source: I have a few 10gbase ones and related mikrotik products, it was a mistake, as in not worth the trouble and costs.

But its fascinating that some public sources like FS communicate like 25GBASE-T is some up and coming sought solution.

Even if you build your copper interconnect to spec, you are still limited to 30m max. And that’s if you have native rj45, or whatever spec complaint connector is there for cat8, on both switch and nic side, due to necessary transmission power levels.

Effective reach will likely be much lower if QSFP port are used, same situation with much more lenient 10GBASE-T ( up to 100m with rj45 on both sides, 30m max with SFP+ due to power limitation of SFP+ interface, cooling and cooking issues notwithstanding).

So to recap 25GBASE-T and 40GBASE-T:

  • requires shielded and up to spec CAT8 cabling
    • maximum 60m reach theoretically, less if termination uses industry common qsfp modules
    • => datacenter and very short reach only
    • implementation necessitates pulling new cabling, and new cabling cannot be reused in case of upgrade to 100 GBE+ (no potential future reuse versus fiber)
  • 10GBASE-T is energy inefficient per port (2-10x times more per port than fiber) when compared to fiber implementation, 25GBASE-T will be even worse due to basic physical fundamentals (higher transmission frequency, cat 6 → cat 8 is 600Mhz → 2000Mhz)
    • higher OPEX vs fiber, guaranteed not to improve in future, again physics
  • increased power use implies increased cooling cost, additional increase of opex vs fiber
  • 10GBASE-T has order of magnitude higher latency compared to fiber, same will apply to 25GBASE-T and 40GBASE-t

It’s literally no brainer to avoid copper implementation, since its superior solution from all angles of attack.
Opex is the king after all and fiber is not expensive at all.
It seems like consortium finalized specs just in case there is some usecase they arw not seeing, but in the end nobody found it worthwhile to productive due to market realities.

Subnotes - 10GBASE-T vs SFP+ fiber:
0 vR2mJtw8iZl8MSB4
src: 10G Technology: 10GBASE-T Vs. SFP+ | by Emily Twain | Medium

TLDR: 25GBASE-T and 40GBASE-T seems to be both technological and economical deadend compared to fiber implementations. No wonder they exist only on paper.

2 Likes
6

If I recall, Alpha actually soft launched a 25G base-t switch a couple years ago but there were problems with the SPF28>RJ45 transceivers that were required on the NICs (“native” 25G base-t NICs don’t seem to exist).

1 Like
7

No products currently exist, but it will be interesting to see where products go in the next 5-10 years. The top spec of wifi 7 should be able to get a real world throughput of 5gbps or more (6E already hit 2.5gb real world on my phone and wifi 7 has 2x channel width and more efficiency), and so wifi 8 will likely push this to 10gbps real world over wifi. At that point, where do homes go? Will someone actually start making 25GBASE-T products, or will homes finally be forced to migrate to fiber?

8

Okay so there’s 1 product at least announced, Alpha Networks switch SNC-60x0-486T: 最新消息:Upgrading from 10G BASE-T to 25G BASE-T is the current trend The Alpha Networks debuts the first 48-port 25G BASE-T Ethernet switch in the world-Alpha Networks Inc.

And, I guess, Foxconn makes RJ45 jacks for 25GBase-T: https://www.fit-foxconn.com/mainssl/modules/MySpace/BlogInfo.php?xmlid=19012 It’s telling that it looks like when they demonstrated the switch at DesignCon they had the switch plugged into … itself. I’m sure it’s selling like hotcakes.

FIT _ Alpha switch setup to test 25G BASE_T
FIT _ Alpha switch setup to test 25G BASE_T1065×760 84.5 KB

1 Like
9

WORLDS FIRST 25 GIG COPPER SWITCH MANAGED BY WORLDS FIRST WINDOWS 7 LAPTOP.

kekw

3 Likes
10

Someday we’ll presumably need it.

The main issue with fiber is a low insertion cycle rating (the number of times you can connect/disconnect). Fantastic for stationary semi-permanent stuff. I have fiber in my home where I can use it. But mobile devices (say laptops) just aren’t very amenable to a direct fiber connection. I haven’t seen a laptop with SFP+ or fiber built in, of course. My workstations and desktops have 10GbaseT on board. Racked general purpose gear has DAC or fiber via PCIe cards. ‘Trunk’ runs are fiber. Laptops get to 10GbaseT via Thunderbolt (Sonnet or CalDigit). Switches have SFP+ and 10GbaseT. No SFP28 yet.

I see PoE as baseT’s main long-term upside in permanent installations. Access points, cameras, IoT, etc. It’s almost nuts how much power you can push to PoE++ devices, and you can manage it remotely. Not to mention less wiring. But we haven’t really had real market demand for 10G PoE, and even wireless access points were slow to adopt 2.5G. Do we one day wind up with the existing twisted pair plant being relegated to power delivery and not communication? What replaces it?

I’d love to go to fiber everywhere I don’t need PoE. I don’t see that happening anytime soon. But I don’t relish the thought of having to run separate power to access points and cameras runs in my home. :slight_smile:

1 Like
11

Yeah, spurred on by this and by how cheap fiber patch cables are now I was daydreaming about a possible consumer-grade optical networking standard recently. It would have to have more robust connectors and cables, as you mention. It would probably need to have transceivers smaller than SFP+. It doesn’t look like those transceivers are too packed so it should be possible. It should probably support some kind of data rate auto-negotiation. It doesn’t seem particularly far-fetched.

Of course, Intel tried pretty hard to make consumer optical cables work with Thunderbolt (nee Light Peak) and couldn’t make it work. That was over ten years ago, though, and they didn’t need cables much longer than a meter for their design use-cases.

The main roadblock is probably not technological but a lack of a market. How many consumers really need very high speed wired data transfer? How many need it and can’t make do with enterprise networking gear or short thunderbolt cables?

12

Ooooh… This takes me back to my early twenties, when I was studying for taking the CCNA certification (never happened due to dotcom shenanigans, but that is a story for another time). Anyway, I was doing a mock exam and go the question whether a K-mart store that only sends the salary books every week should invest in DSL or 56k dialup.

Me, totally confident and from a country with good internet connections, of course answered DSL, duh! With dialup costing $2.50 an hour and DSL being $20 a month it was a no brainer!

The mock exam failed me instantly, because what I did not know was that in the US and at the time the question was written, dialup was 50 cents an hour and DSL started at 80 bucks. So in my country it was $10 for 4 hours of connectivity vs $20 for 720 hours of connectivity, in the US it was $2 for 4 hours vs $80 for 720 hours. At those prices, dialup would indeed be the better option here.

But what neither the questioner nor me knew at the time, was that 3 years later dialup would be gone from most parts o the world. And that is my point here; 10 GbE may be good enough today but it is a dead end and as tech advances, consumer fiber will be more and more important. Wiring fiber everywhere inside the walls should really be the norm going forward.

I did have an idea of a PoE powered fiber switch that fits in a wall socket, has fiber on the back end and ethernet ports on the front. That would be consumer grade enough to at least jumpstart fiberoptic wiring in the walls. Once the fibers are in the walls it is a lot easier to upgrade the outlets.

Oh well, let us see if that ever become a reality :slightly_smiling_face:

13

25GBASE-T is vaporware. Twisted pair is dead beyond 10GBASE-T, and NBASE-T bought a lot of time for good old Cat 5e.

The cabling vendors love pushing it since they can con users into buying over-specced cables but there has not been a single IC with it ever made. It’s useless for datacenter since DAC copper is much lower power, and if you need 25G you can afford to lay down SMF and never have to worry about cabling again

1 Like
14

Can we even say it is dead, if it was never alive in first place?

2 Likes