One of the parts that was quietly mentioned in yesterday’s Cascade Lake X-Series processor launch was support for new Ethernet controllers. Specifically it mentioned Intel’s i225 Ethernet controller range, which are rated for 2.5 GbE networking speed. This is news to us – up until this point, Intel has been relatively quiet on its ‘multi-gigabit’ Ethernet strategy (2.5GbE and 5GbE) and other companies have taken the lead.

Intel’s RJ-45 networking strategy for the last decade has been pretty simple: 1 GbE controllers everywhere, and 10 GbE controllers for enterprise, and some nice margin on the enterprise stuff. The only major update in the 10 GbE strategy in the last few years has been the upgrade from the X540 line to the X550, which moved the PCIe interface from PCIe 2.0 to PCIe 3.0. While other companies have been experimenting with multi-gigabit Ethernet solutions, Intel has been rather quiet.

The market leader in 2.5 GbE and 5 GbE controllers so far has been Aquantia, with their AQC107 and AQC108 chips (plus derivatives). Aquantia was acquired by Marvell earlier this year, and we expect those product lines to be rolled up into Marvell’s naming schemes. Aquantia has had good success with multi-gigabit Ethernet for backhaul connections and automotive, so it will be interesting to see if Marvell will keep the retail channel alive. Aquantia has been the market leader in this space for three years.

The other entrant is Realtek, who released a consumer 2.5 GbE controller earlier this year. This is getting some traction on some consumer motherboards, but is still an outside play – most OEMs are looking at Aquantia’s solution for now. We also have Rivet Networks’ E3000 solution, announced back at CES 2019.

The main barrier to 2.5 GbE and 5.0 GbE adoption has ultimately been the lack of consumer grade multi-port switches. We’ve seen some devices offer single or dual ports up to 2.5 GbE/5 GbE/10 GbE, but these can’t be daisy chained or are limited to the equivalent of peer-to-peer direct connection. Users wanting something bigger, like I did, have to look to the commercial space, such as the XS724EM which more of a multi-user office 24-port switch and costs a pretty penny.

With Intel now set to enter the market with its own 2.5 GbE controller offering, we can hope it means several things. Firstly, that Intel doesn’t see this as a threat to their 10 GbE revenues. Second, hopefully switch manufacturers take the same view, and we see more multi-gigabit Ethernet switches for mainstream customers. Thirdly, we hope that they will be widely available for motherboard manufacturers to use. Given concerns about Intel’s recent 14nm demand issues, it will be interesting to see at what rate these controllers are manufactured.

When speaking with Intel, we were told that availability of the i225 family for its OEM customers will occur later in Q4, as the company ramps up production. So we’re likely to see more of these ports in 2020.

Intel’s ARK database has pages for its i225 controller family. At the time of writing, they incorrectly state that the i225 is a 1 GbE controller – we can confirm they are 2.5 GbE controllers.

Related Reading

 

Comments Locked

31 Comments

View All Comments

  • azazel1024 - Friday, October 4, 2019 - link

    Other than a broken or improperly terminate cable, I've never seen a performance issue over a "bad cable". By broken I mean exactly that, one of the wires is actually broken. Improperly terminated I mean one or more of the wires is not contacting the terminal connector. That results in it falling back to fast Ethernet, which is possible over 2 pair, rather than 4 pair.

    Now, I've probably only installed a few hundred cables and tested a couple times that. I am sure something is always possible. I've never installed in a very high EMI environment and when I run cables I don't bundle dozens together.

    If I have quality NICs on both ends, I've never failed to get darn close to 1000Mbps after overhead is accounted for a 1GbE links.

    My reluctance to upgrade is the cost. I want and can leverage 2.5GbE on my network. But I don't want to spend $200-300 on a core switch that only has 8-10 total ports, with 2 of them being 2.5GbE. I'd want a switch that has a minimum of 6 ports capable of 2.5GbE and ideally it would have 8+ ports capable of 2.5GbE and have a minimum of 16 total ports (24 would be better) with the rest being 1GbE (all ports being 2.5GbE+ would be nice, sure).

    My cost limit is around $200-300 for such a beast. Then it is the NICs, I'd be okay starting out with only a single port NIC with a 2.5GbE port in my server and my desktop, but again, I wouldn't want to pay more than maybe $100 for each NIC. Call it $500 max to upgrade the basics of my network. Down the road add another NIC to each machine when they are cheaper or I need or want the extra bandwidth.

    But I don't need/feel like I need to replace my switch in a year or three if I end up with more than 2 ports requiring 2.5GbE. Or have some kludge where I need multiple switches networked together to handle my home network. I am fine running a cable to a little 5 port so that my AppleTV, Xbox and some other thing can all be on the network at my TV rather than running 2 or 3 or 4 cables over there. But I don't need my "core network" to have to have multiple switches to handle all of the drops and all of the devices on my network.

    We aren't there yet, but the last few months have been a lot of hopeful news. It does feel kind of on the cusp of having "affordable" 2.5GbE networking.
  • Kevin G - Friday, October 4, 2019 - link

    There are some good industrial RJ45 connectors. In fact, the ones I've been using are actually re-usable (both male and female RJ45). The downside is that each connector is several US dollars in bulk.

    But you are correct, a mismatched RJ45 connector and end can be a nightmare. I recall taking an hour on one termination as the CAT6A wire gauge was mismatched with the shielded connector.

    There is a replacement for RJ45 called GG45 which is aiming to be the connector for 25 GbaseT and 40 GbaseT interfaces. The unraveling of the twisted pairs to make a shielded connector causes problems at these data rates so the pins for two pairs move to the top of the connector. Interestedly, the GG45 profile is the same as RJ45 and pseudo backwards compatible as 100 Mbit speeds can be had with a GG45 cable in an RJ45 socket (or vice versa). This also opens up the possibility of yet another connector but this time being 12c12p with the same RJ45 profile but compatible with both GG45 and RJ45 cabling as well as its own. I wouldn't want to work with such a cable as that would have to be incredibly stiff and a super pain to terminate.
  • Givmedew - Friday, October 4, 2019 - link

    You are wrong. In the home environment almost all cabling is 10G capable. I've been installing Cat6 or better for 20 years. I've never encountered worse than 5e and 5e is rated at 45m for 10GBit/s.

    That's the rated distance on 5e but personally I've seen it fail 1GBit without replacing the terminations with shielded terminations.

    After terminating again I have not encountered very much that couldn't do 10Gbit. I would expect 2.5gbit to be rated at better distances since it will have better newer noise resistance technology as well as it being slower. The 10G standard is super old. Using some of the technology present in Wireless6 should surely allow pure crap copper that can't even get a 1Gbit connection to possibly do better.

    The problem is who wants this? I physically have met 4 people with higher than 1Gbit home networks. I'm including myself. My 2 best friends have nearly identical systems to myself since co developed and split research. I spent nearly 200HRs over 3yrs, my friend is in the business and designs these systems still had to invest over 100HRs of research on how to get a $40,000-100,000 5-10 year old system running for under $1000 per person not including any drives over 2TB and not including NVMe storage for virtual machines. I'm using 30 2TB drives and that is in my budget. All of those systems have bonded 10g or a single 40g on the storage and then up to 3 or 4 10g PCs in the house. I did new runs so I used fiber because it's rated at 100g and I could do 40g in the future.

    My friends house was wired in Cat6. None of those connections would run at 10g without terminating with shielded ends.

    His office we decided to just duck tape fiber to the end of the Cat6 and pull it on through. You can do that in a house.

    Another reason why I call bullshit for bad cables.

    Businesses can't upgrade from 1G to 10G without considerable investment if they are on Cat5e or Cat6.

    100m runs are nothing. You gotta re-run everything with Cat7.

    So it has nothing to do with the enterprise market either. With 10g being so outdated that I can pick up a 4 port 10g with 48 ports of 1g for under $100... Enterprise has already moved past 1g EXCEPT for offices.

    I say holding back had nothing to do with enterprise. But enterprise could be a new income source for 2.5g especially if it utilizes better noise rejection. If you can get companies with old copper to upgrade to 2.5g then maybe it makes sense.

    But realistically my money is on home internet being the real reason we have never seen an increase over 1g. I've been using 1g or better for well over a decade. But ONLY for moving video over Ethernet. Now that I finally have 500mbit internet well 1Gbit is the minimum home speed. My first dual gigabit motherboard was probably in 2004. Bonded (enterprise router) it could see 120MByte/s on 2 streams. That was the near max speed of a hard drive at the time. 2 drives in raid could do 200-300MByte. (Eventually bonded Ethernet could transfer single files at bonded speeds and nowadays you can fully utilize bonding across huge connections even for single file transfers)

    But no one else I knew utilized 1gigabit. My first really fast internet was Verizon FiOS in 2008 at maybe 50mbit if I remember. Since I left FiOS land my first internet that required a gigabit was in mid 2013. I got the call from Comcast that they upgraded my area to 125mbit and in real life Comcast would let you burst close to 200mbit.

    Now people needed gigabit. But wireless was already exceeding 300mbit in real use cases and 100mbit was no problem.

    2017 I move to an area that has gigabit speeds for under $100. Since they are running an older system that maxes out at 50mbit up I decide to buy 500mbit down 50 up since 1000/50 was $20 more.

    I don't see 2Gbit being available for at least 5 if not 10 years. It depends on Google and Comcast. I have WOWay (Wide Open West). It could come to market in 5 years for Greater Chicagoland if Google's sneaky plan to infiltrate Chicago works. But my money is on 10 years.

    So they have 5-10 years to get that tech into routers.

    Most people won't care!!!

    Real world Wireless 6 speeds on 2.4GHz exceed 1Gbit. 4.8GBit on 5GHz band. You'll want your main PC if they exist to be connected to Ethernet if you are a pro-sumer/enthusiast. Nobody else will care.

    Wireless 5 really does exceed 500mbit. I get 600mbit network test and 550mbit speed test on my 5GHz WoW provided router. Disabled in bridge mode I use Meraki Cloud Mesh and see around 100mbit on 2.4GHz with older Meraki units. I could double or triple that but am waiting for the license to expire and will install Wireless 6 APs.

    With wireless 6 we don't need anything better until home internet gets to 10Gbit. Even then... With individual users seeing 1 to 5gbit speeds I doubt end users will want faster. At these speeds we really are talking about the best NVMe SSD speeds over the internet.

    So who needs faster than 1Gbit?

    For now? Nobody. You can already do 2Gbit file single file transfers with bonding. Actually I have 1 computer that is quad bonded and it will do over 3gbit/s I see 350MByte file transfer speeds. It's not really fast enough for video editing. 10GBit is the minimum for 4k pro res at 756GByte per hour or 12 minutes per hour to transfer from the mixer to your work station. Right now not even video editing must have 10Gbit because for a lower invest you can put an NVMe drive into an enclosure and get 8-12Gbit write and 15-20Gbit read from a single external NVMe thunderbolt enclosure. You can get 20gbit write and 40gbit read on multi NVMe enclosures.

    They cost less than buying new 10G and 40G network equipment. So this is yet another hurdle in the pro-sumer, enthusiast, small business sector.

    When moving video in an office do I need to use a network? NO

    So again I go back to the Internet. Internet is going to drive connection speeds.

    To my knowledge you can NOT utilize bonding for IP without some serious hardware and anyways you can't bond connections on the routers you get from the internet company.

    When people can buy internet that is over 1gigabit Intel or someone else will ALREADY need a solution in place.

    Similar to 1g being in motherboards 10 years before it was needed. 2.5g needs to be in motherboards before anyone can use it. My guess is there will never be a large retail market for 2.5g by the time it hits shelves the router you rent will have 2.5g and absolutely no one is going to need more than the 4 ports typically included. Once 5g home internet comes along 5g Ethernet will need to be ready for routers. But 2.5 and or 5g Ethernet will need to be ready soon for motherboards.

    Lastly some 10g motherboards exist. If Intel makes 2.5g and 5g available they protect their 10g enterprise by killing low cost 10g consumer products dead in the water!

    My dual 10g dual 1g Mellanox card probably cost some business $1000. I paid $50 from a company selling 1000 used ones. They probably upgraded from bonded 10g (20g in certain scenarios) to 100g.

    Also creating 2.5g and 5g creates an important segment for Intel.

    If you go to eBay you can buy used server crap for nothing. I picked up 2 dual 40g Mellanox cards for $10/ea. I haven't used them but I can put them into my server and my main PC and direct connect them for up to 80Gbit of connection.

    Used 2.5g and 5g stuff doesn't exist.

    This is my analysis of this product and it comes from 20 plus years experience utilizing enterprise stuff at home. I had my first managed enterprise switch when I was 14 in 1997. It had 10mbit 100mbit auto-sensing could do bonding and importantly plugging in a friend who had 10mbit didn't knock everyone else down to 10mbit (common back then). I used it for LAN parties.
  • abufrejoval - Thursday, October 3, 2019 - link

    The switch situation has markedly improved IMHO: I started with the ASUS XG-U2008, which was wonderfully quiet and smal, but only offers 2 10Gbase-T ports and no NBase-T support (2.5/5Gbit) and then added a Buffalo BS-MP2012 true NBase-T switch, which was certainly cheap enough at ~€50/port if a bit noise without hacking.

    For me noise is a crucial issue, because few 10Gbit switches are built for office or home acoustics. I have workstations running almost inaudible when pushing 800Watts out of the chassis while a 45Watt NBase-T switch can make a room impossible to work in.

    The hacking (low-noise constant flow Noctua fans) might have bit me on the Buffalo, because the other day one of the three switch-side Aquantia 417 might have gone bad: I lost four ports out of 12 so I went a-hunting for a replacement, which would perhaps be quiet without the hacking.

    I got a Netgear XS508M 8-port variant, and that turns out to be wonderfully quiet. Yes, it has a fan and perhaps things would get a little noisier once I were to connect lots of older Intel 10Gbase-T cards without energy efficient Ethernet support, but so far the fan is next to inaudible with 2-3 ports connected running iperf3 for a good 10 minutes or so.

    And those Netgears come a quite a few sizes (4,8,12,24 port) and pretty linear prices, somewhat comparable to what the NICs cost, too.

    I have been using Aquantia AQC107 10Gbit where I have the slots, RealTek 2.5Gbit USB3 for Gemini-Lake µ-Servers and notebooks and even the QNAP 5Gbit USB3 for a Xeon-D that unfortunately lacks the PHYs for the native 10Gbit ports and needs its PCIe x16 slot for a GPU.

    After being stuck with Gbit Ethernet for so many years, in this day and age of affordable SSD storage on home-servers, my blood pressure is no longer spiking on larger file copies and backups.
  • abufrejoval - Thursday, October 3, 2019 - link

    sigh, sorry for those typos, need edit...
  • yeeeeman - Friday, October 4, 2019 - link

    The announcement was pretty loud, heard it from 5000 miles away. Joking.
    Given the fact that 1Gbps is more than enough for 99% of consumers, I still see no point yet in moving to 2.5Gpbs silicon. It is more efficient to carry less data (i.e. compress/decompress data), than to up the speeds. I have been on a 300mbps connection for 10 years now. I have never ever felt I needed more than that. Sure, would be nice to finish an 100gb game download in 2 minutes, but while it downloads I can watch some youtube, no hurry.
  • dromoxen - Friday, October 4, 2019 - link

    This would suit me BUT I would need a new mobo for it..1Gb is nice but 2.5Gb would make a good difference, this between two PCs @ home (video in/out). Wireless does eem to be leading the pack in R+d, and is the way of the future , except for hacking/sniffing worrys
  • yetanotherhuman - Friday, October 4, 2019 - link

    I look forward to it. I find the Aquantia cards kind of unstable, but to be fair, I haven't upgraded them to the newest firmware yet.
  • Diogene7 - Friday, October 4, 2019 - link

    I have some questions :

    1. Would it be possible to re-architect 10GbE or more using USB-C port (ex: a future Thunderbolt 4.0 standard as Thunderbolt 3 already have a theoritical bandwith of 40Gbit/s) ?

    2. From there, as USB-C Thunderbolt 3 can theoritically deliver up to 100W power, would it be possible to develop switches PoE (Power over Ethernet) USB-C switches that deliver up to 100W power

    3. Finallly, would it be possible to integrate those 100W USB-C port instead of USB-A port that we sometimes find on some modern power socket ?

    The goal would be to unify an architecture that distribute up to 100W power by USB-C socket, while also being a network backbone for an entire house...
  • Azlehria - Saturday, October 5, 2019 - link

    The problems with USB as a network bus cable are manifold. First, USB cable, and plugs, are expensive. Second, it's effectively impossible to field-terminate. Third, although they're rated for more insertion/removal cycles than typical 8P8C connectors, USB jacks and plugs tend to be less resiliant when abused. Fourth, testing equipment is a niche product that's difficult to find. Fifth, any power circuit larger than a single 20 AWG pair is not compliant with the USB spec, which designs for a maximum cable length of 3 meters. Sixth, that pair is de facto guaranteed to be stranded, reducing its capacity over distance.

    Once you reach horizontal cabling lengths, fiber is cost-competitive with USB and *can* be terminated in the field. It's not, effectively, distance-limited. Testing and even certification equipment is readily available, even rentable. You can't run power over fiber, but PoE is often incompatible with existing TP plant due to the changes in conduit fill limits. Whereas you can replace TP with fiber and run mains power in the same conduit - fiber is non-conductive and non-heating, so most building codes allow it to be run in the same raceways as mains cable, unlike other communications cable.

    On the other hand, if your existing plant allows or if you're installing new plant, 4PPoE permits power delivery up to 100W already, although I'm not sure any products are on the market yet as 802.3bt was only finalized last year. Standard PoE+, which is widely implemented, allows up to 25.5W, and non-standard PoE implementors such as Cisco, Ubiquiti, et al. have long supported higher power limits up to at least 112W in some installations.

Log in

Don't have an account? Sign up now