• Win 11 PRO with 2.5Gbps ethernet

    From Jeff Barnett@jbb@notatt.com to alt.comp.os.windows-11 on Sun Apr 5 01:45:16 2026
    From Newsgroup: alt.comp.os.windows-11

    I recently upgraded my LAN by replacing wired 1Gbps switches with
    TRENDnet 5-Port 2.5Gb Switches (TEG-S351, 5 x 2.5G RJ45 Ports, 25Gbps Switching Capacity). Two of my machines are built from Win 11 PRO on MSI
    MPG Z790 Carbon WIFI II motherboards, Intel I9-12900K processors, 65GB
    memory, good video cards, and SSD storage. (I mention the configuration elements to simply make the point that these computers are adequately
    power for most tasks.) After the aforementioned upgrade, I noticed that
    the computer to switch wires were still running at 1Gbps.

    I rebooted these machines but the low speed stuck. Next, I looked in the
    BIOS to see if I had overlooked some setting there. That wasn't it
    either. Finally, I looked in the device manager and consulted the
    advanced tab on the properties of the Intel Ethernet Controller listed
    under Network adapters. I found a relevant setting there: Speed and
    duplex where the following choices were offered: 10 Mbps Full Duplex,10
    Mbps Half Duplex, 100 Mbps Full duplex, 100 Mbps Half Duplex, 2.5Gbps
    Full Duplex, Auto Negotiation. The value used for 1Gbps was Auto
    Negotiation; changing to 2.5Gbps fixed my problem.

    If you have read this far, first thank you. Now I have some questions:

    Q1: What standard or reasoning keeps Auto Negotiation from picking up
    the 2.5Gbps option? Is there a standards change in the pipeline to cure
    this oddity?

    Q2: Do both 1Gbps and 2.5Gbps imply full duplex by standards or
    convention? I noticed that the switches I'm using specify their capacity
    at 25Gbps, i.e., at 5Gbps for each of the 5 sockets on the switch.

    Q3: Is the peculiar set of choices offered a M$ artifact, other than
    they probably participated in the authoring standards committee? For
    example do Unix/Linux boxes have the same odd interface to basic net gear?
    --
    Jeff Barnett

    --- Synchronet 3.21d-Linux NewsLink 1.2
  • From Paul@nospam@needed.invalid to alt.comp.os.windows-11 on Sun Apr 5 07:33:44 2026
    From Newsgroup: alt.comp.os.windows-11

    On Sun, 4/5/2026 3:45 AM, Jeff Barnett wrote:
    I recently upgraded my LAN by replacing wired 1Gbps switches with TRENDnet 5-Port 2.5Gb Switches (TEG-S351, 5 x 2.5G RJ45 Ports, 25Gbps Switching Capacity). Two of my machines are built from Win 11 PRO on MSI MPG Z790 Carbon WIFI II motherboards, Intel I9-12900K processors, 65GB memory, good video cards, and SSD storage. (I mention the configuration elements to simply make the point that these computers are adequately power for most tasks.) After the aforementioned upgrade, I noticed that the computer to switch wires were still running at 1Gbps.

    I rebooted these machines but the low speed stuck. Next, I looked in the BIOS to see if I had overlooked some setting there. That wasn't it either. Finally, I looked in the device manager and consulted the advanced tab on the properties of the Intel Ethernet Controller listed under Network adapters. I found a relevant setting there: Speed and duplex where the following choices were offered: 10 Mbps Full Duplex,10 Mbps Half Duplex, 100 Mbps Full duplex, 100 Mbps Half Duplex, 2.5Gbps Full Duplex, Auto Negotiation. The value used for 1Gbps was Auto Negotiation; changing to 2.5Gbps fixed my problem.

    If you have read this far, first thank you. Now I have some questions:

    Q1: What standard or reasoning keeps Auto Negotiation from picking up the 2.5Gbps option? Is there a standards change in the pipeline to cure this oddity?

    Q2: Do both 1Gbps and 2.5Gbps imply full duplex by standards or convention? I noticed that the switches I'm using specify their capacity at 25Gbps, i.e., at 5Gbps for each of the 5 sockets on the switch.

    Q3: Is the peculiar set of choices offered a M$ artifact, other than they probably participated in the authoring standards committee? For example do Unix/Linux boxes have the same odd interface to basic net gear?

    MDI/MDIX was introduced around 1GbE timeframe. It allows the
    four pairs used for GbE, to work with either a straight
    cable or with a crossover cable (the yellow cable in the
    router box). Now, was that standard forward looking
    and assume any faster speeds would automatically be tested ?
    Dunno.

    The four pairs, might use PAM5 and are bidirectional.

    https://www.edn.com/what-pam5-means-to-you/

    A GbE NIC can do GbE in both directions at the same time.
    The person answering here, says an echo canceller allows
    driving both ends of the wire at the same time. Using your
    own waveform, you can distinguish when the other end is
    driving towards you.

    https://electronics.stackexchange.com/questions/132486/how-does-bidirectional-transmission-on-gigabit-ethernet-work

    As the rate goes up, eventually the link encoding has to
    change. Like it does for all the other low-amplitude standards.

    The PAM5, was so the signal occupied the same power spectrum
    as a previous standard. Presumably this allows the same
    cable type to be used for both. Once you mess around
    and change a lot of stuff, to achieve greater speeds, eventually
    that is reflected in newer cable standards.

    *******

    None of that is answering your question. To me, this isn't
    a Windows versus Linux issue, it's just the usual sloppy
    hardware design with the usual sloppy software design,
    placed on a plate for the user to eat. You do whatever
    is necessary to get your jalopy working.

    There is at least one higher-speed NIC, where the
    hardware people decided to drop some of the modes,
    to reduce the work they had to do. Some of the chip
    designers, they just buy an IP block for the function
    and part of that can be a "certified PHY" so a lot less
    high speed scoping and verification is needed.

    I've got some cards installed, which are currently disabled,
    as manually using the cards causes some networking problems
    and I will need at some point, to acquire a higher speed
    switch to complete my project. I'm waiting for the RealTek
    chips. There is a version they were supposed to have made
    for switch and router designs, so that a lower cost
    switch will be made available. Whether that will happen,
    who can say these days. I thought when RealTek announced
    their stuff, we would see it flood into the stores. So far,
    it has not looked like that.

    Paul
    --- Synchronet 3.21d-Linux NewsLink 1.2