• very slow download speeds

    From Dan@3:633/10 to All on Sunday, June 14, 2026 18:02:50
    Hello,

    I am on Community fibre, symmetrical 2.5 gbit. On Nperf I get 1.7 down
    and 2.4 up.
    Upload is 2.4 but download has decreased to 700 Mbit. I am using Ookla
    app in Win 11.
    My Ethernet is a TP link TX 401 10 Gig. Cables, all are Cat 8. Tried
    different ones with the same result. My ONT is 10 Gig capable.
    Got the latest driver from TP link, installed it and no difference.

    Tested with Linux mint and I get 2.4 down and up.

    --- PyGate Linux v1.5.16
    * Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)
  • From ....winston@3:633/10 to All on Sunday, June 14, 2026 17:01:45
    On 06/14/2026 1:02 PM, Dan wrote:
    Hello,

    I am on Community fibre, symmetrical 2.5 gbit. On Nperf I get 1.7 down
    and 2.4 up.
    Upload is 2.4 but download has decreased to 700 Mbit. I am using Ookla
    app in Win 11.
    My Ethernet is a TP link TX 401 10 Gig. Cables, all are Cat 8. Tried different ones with the same result. My ONT is 10 Gig capable.
    Got the latest driver from TP link, installed it and no difference.

    Tested with Linux mint and I get 2.4 down and up.

    Router between fiber and device?
    Device using wifi or cable?
    Device speed test performed on wifi or cable
    Device's network card/adapter specs?
    Using a VPN? If so, disable and retest.

    CAT8 is usually overkill, but acceptable.

    For Windows, is TCP auto-tuning enabled?
    Run this command in a Powershell admin:
    netsh interface tcp show global

    What does Auto-tuning show?
    - A 'Normal' result = Enabled

    As a test, if Normal, disable and retest
    netsh interface tcp set global autotuninglevel=disabled
    Restart the device, then retest

    Note Auto-tuning can be turned back on:
    netsh interface tcp set global autotuninglevel=normal


    --
    ...wнд?ё?д

    --- PyGate Linux v1.5.16
    * Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)
  • From Dan@3:633/10 to All on Monday, June 15, 2026 07:42:33
    On Sun, 14 Jun 2026 17:01:45 -0400, "....winston"
    <winstonmvp@gmail.com> wrote:

    On 06/14/2026 1:02 PM, Dan wrote:
    Hello,

    I am on Community fibre, symmetrical 2.5 gbit. On Nperf I get 1.7 down
    and 2.4 up.
    Upload is 2.4 but download has decreased to 700 Mbit. I am using Ookla
    app in Win 11.
    My Ethernet is a TP link TX 401 10 Gig. Cables, all are Cat 8. Tried
    different ones with the same result. My ONT is 10 Gig capable.
    Got the latest driver from TP link, installed it and no difference.

    Tested with Linux mint and I get 2.4 down and up.

    Router between fiber and device?
    Device using wifi or cable?
    Device speed test performed on wifi or cable
    Device's network card/adapter specs?
    Using a VPN? If so, disable and retest.

    CAT8 is usually overkill, but acceptable.

    For Windows, is TCP auto-tuning enabled?
    Run this command in a Powershell admin:
    netsh interface tcp show global

    What does Auto-tuning show?
    - A 'Normal' result = Enabled

    As a test, if Normal, disable and retest
    netsh interface tcp set global autotuninglevel=disabled
    Restart the device, then retest

    Note Auto-tuning can be turned back on:
    netsh interface tcp set global autotuninglevel=normal





    Microsoft Windows [Version 10.0.26200.8655]
    (c) Microsoft Corporation. All rights reserved.

    C:\Windows\System32>netsh interface tcp show global
    Querying active state...

    TCP Global Parameters
    ----------------------------------------------
    Receive-Side Scaling State : enabled
    Receive Window Auto-Tuning Level : disabled
    Add-On Congestion Control Provider : default
    ECN Capability : disabled
    RFC 1323 Timestamps : allowed
    Initial RTO : 1000
    Receive Segment Coalescing State : enabled
    Non Sack Rtt Resiliency : disabled
    Max SYN Retransmissions : 4
    Fast Open : enabled
    Fast Open Fallback : enabled
    HyStart : enabled
    Proportional Rate Reduction : enabled
    Pacing Profile : off


    C:\Windows\System32>






    Router between fiber and device? Cat 8 Ethernet
    Device using wifi or cable? Cat 8 Ethernet
    Device speed test performed on wifi or cable Cat 8 Ethernet
    Device's network card/adapter specs? TP link TX 401 10 Gig 10 Gig
    enabled
    Using a VPN? If so, disable and retest. No VPN



    --- PyGate Linux v1.5.16
    * Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)
  • From Paul@3:633/10 to All on Monday, June 15, 2026 03:55:52
    On Sun, 6/14/2026 1:02 PM, Dan wrote:
    Hello,

    I am on Community fibre, symmetrical 2.5 gbit. On Nperf I get 1.7 down
    and 2.4 up.
    Upload is 2.4 but download has decreased to 700 Mbit. I am using Ookla
    app in Win 11.
    My Ethernet is a TP link TX 401 10 Gig. Cables, all are Cat 8. Tried different ones with the same result. My ONT is 10 Gig capable.
    Got the latest driver from TP link, installed it and no difference.

    Tested with Linux mint and I get 2.4 down and up.


    I have three TX 401, and the testing stopped after I got two installed.
    (I have no 10GbE switch for them, which is why the project is stalled
    waiting for some cheep RealTek based box to "come along".)

    I'm running them on CAT6A. In point to point test, I got around 1000MB/sec instead of a number derived from 10GbE.

    The purpose of selecting the TX 401, is so the PCI Express slot
    does not have to be too fancy. That's so the card could work in
    the 4930K machine. You need enough lanes (it's an x4 card) and
    enough lane speed, to make it to 10GbE. And none of my equipment
    is exactly smoking-fast. On older equipment, you have to take
    the sub-sized buffers into account in the chipset -- the "lanes
    need to be twice as fast as needed" to cover the buffer issue.
    Later equipment seemed to fix that somehow. PCIe has gotten closer
    to wirespeed since the white paper on buffer size versus speed came out.

    There will be some cheaper ones, from RealTek, which run on
    one PCI Express Rev4 lane, but not many machines will be kitted
    out properly for that particular one. Whereas the TX 401 will run
    in more of your machines (maybe even a 10-12 year old machine).
    A second RealTek chip uses two PCIe Rev3 lanes (and presumably on an x4 shaped piece of PCB). Which would not be the best for my 4930K with PCIe Rev2
    lanes a-plenty. I think that's where the card is in that machine, on a
    PCIe Rev2 interface. RealTek does not appear destined to make a
    four lane version (their plan, after all, is to make dirt cheap
    kit, even if it sucks around the edges a bit).

    I don't have a 10GbE Switch yet for the thing, which is why I have
    tested them point to point. And my project is in limbo, as I wait
    for worthwhile materials to buy, to finish the job.

    The TX 401 on this machine, is Code 22 (as I have disabled
    it on purpose (to save power while it is not doing anything).
    Driver is Marvel AQtion 3.1.10.0 Tuesday 4/23/2024 . (What catalog.update.microsoft.com
    seems to be serving, is 3.1.10.0 . Tested this, by installing the third card to see
    what driver is handed out, and that is the auto-installed one.)

    The problem with testing them with "file sharing", is the
    link is encrypted, and that could be a limitation for that
    style of testing. As part of your "transport", there should
    be less of a restriction in the path there.

    https://www.servethehome.com/tp-link-10gigabit-pcie-network-adapter-mini-review-marvell-aqc113/

    # It kinda looks like a CSP (chip scale) package, but maybe it's considered to be a fine pitch BGA
    # Only important if you suspected the drivers for the three entries were somehow different.
    # DO NOT remove the heatsink, unless you know you have the thermal tape or paste suited
    # to the chip. I don't know if it is booby trapped, for example.

    Marvell AQC113-B1-C chip

    https://www.marvell.com/support/downloads.html

    Category = Marvell Public Drivers
    Platform = Windows
    Part Number = AQC113
    Keyword = <blank>

    10/17/25 Marvell AQtion Windows 10 and 11 64-bit Driver
    Marvell AQtion Windows 10 and 11 64-bit Driver
    Marvell Public Drivers
    Windows 10 and Windows 11 64-bit driver, windows-10-64-bit-driver Driver, Drivers 3.1.11

    Name: Marvell_AQtion_Win_v3.1.11_10-17-2025.zip
    Size: 664693 bytes (649 KiB)
    SHA256: 81018243ED078AA3ACCCFD9E88EA189D4A79C8BA8B0998A6EF44AD041B324A78

    One of the folders has this in it:

    Marvell_AQtion_3.1.11.0_installer_win1x.msi 593,520 bytes

    and right-clicking on that should work just as well as using the .INF setup in
    the other folder.

    *******

    Over file sharing, I got 1.03GB/sec while transferring a 7GB DVD between RAM drives.
    So that's file sharing.

    The servethehome got 9000 on iperf, and dividing that by 8 gives 1.125GB/sec
    so my measured value is a little less.

    *******

    The "Random number" of 700Mbit/sec does not ring any bells.

    If you're watching the Ethernet interface in Task Manager, does the
    transfer rate change with time ? My somewhat more modest network
    connection, has a downward glitch every 5 seconds or so, which is
    presumably an issue with the box at the corner. And mine is a wired connection. The plan for the 10GbE cards was purely for transfers between machines.

    Paul

    --- PyGate Linux v1.5.16
    * Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)
  • From Dan@3:633/10 to All on Monday, June 15, 2026 12:37:40
    On Mon, 15 Jun 2026 03:55:52 -0400, Paul <nospam@needed.invalid>
    wrote:

    On Sun, 6/14/2026 1:02 PM, Dan wrote:
    Hello,

    I am on Community fibre, symmetrical 2.5 gbit. On Nperf I get 1.7 down
    and 2.4 up.
    Upload is 2.4 but download has decreased to 700 Mbit. I am using Ookla
    app in Win 11.
    My Ethernet is a TP link TX 401 10 Gig. Cables, all are Cat 8. Tried
    different ones with the same result. My ONT is 10 Gig capable.
    Got the latest driver from TP link, installed it and no difference.

    Tested with Linux mint and I get 2.4 down and up.


    I have three TX 401, and the testing stopped after I got two installed.
    (I have no 10GbE switch for them, which is why the project is stalled
    waiting for some cheep RealTek based box to "come along".)

    I'm running them on CAT6A. In point to point test, I got around 1000MB/sec >instead of a number derived from 10GbE.

    The purpose of selecting the TX 401, is so the PCI Express slot
    does not have to be too fancy. That's so the card could work in
    the 4930K machine. You need enough lanes (it's an x4 card) and
    enough lane speed, to make it to 10GbE. And none of my equipment
    is exactly smoking-fast. On older equipment, you have to take
    the sub-sized buffers into account in the chipset -- the "lanes
    need to be twice as fast as needed" to cover the buffer issue.
    Later equipment seemed to fix that somehow. PCIe has gotten closer
    to wirespeed since the white paper on buffer size versus speed came out.

    There will be some cheaper ones, from RealTek, which run on
    one PCI Express Rev4 lane, but not many machines will be kitted
    out properly for that particular one. Whereas the TX 401 will run
    in more of your machines (maybe even a 10-12 year old machine).
    A second RealTek chip uses two PCIe Rev3 lanes (and presumably on an x4 shaped >piece of PCB). Which would not be the best for my 4930K with PCIe Rev2
    lanes a-plenty. I think that's where the card is in that machine, on a
    PCIe Rev2 interface. RealTek does not appear destined to make a
    four lane version (their plan, after all, is to make dirt cheap
    kit, even if it sucks around the edges a bit).

    I don't have a 10GbE Switch yet for the thing, which is why I have
    tested them point to point. And my project is in limbo, as I wait
    for worthwhile materials to buy, to finish the job.

    The TX 401 on this machine, is Code 22 (as I have disabled
    it on purpose (to save power while it is not doing anything).
    Driver is Marvel AQtion 3.1.10.0 Tuesday 4/23/2024 . (What catalog.update.microsoft.com
    seems to be serving, is 3.1.10.0 . Tested this, by installing the third card to see
    what driver is handed out, and that is the auto-installed one.)

    The problem with testing them with "file sharing", is the
    link is encrypted, and that could be a limitation for that
    style of testing. As part of your "transport", there should
    be less of a restriction in the path there.

    https://www.servethehome.com/tp-link-10gigabit-pcie-network-adapter-mini-review-marvell-aqc113/

    # It kinda looks like a CSP (chip scale) package, but maybe it's considered to be a fine pitch BGA
    # Only important if you suspected the drivers for the three entries were somehow different.
    # DO NOT remove the heatsink, unless you know you have the thermal tape or paste suited
    # to the chip. I don't know if it is booby trapped, for example.

    Marvell AQC113-B1-C chip

    https://www.marvell.com/support/downloads.html

    Category = Marvell Public Drivers
    Platform = Windows
    Part Number = AQC113
    Keyword = <blank>

    10/17/25 Marvell AQtion Windows 10 and 11 64-bit Driver
    Marvell AQtion Windows 10 and 11 64-bit Driver
    Marvell Public Drivers
    Windows 10 and Windows 11 64-bit driver, windows-10-64-bit-driver Driver, Drivers 3.1.11

    Name: Marvell_AQtion_Win_v3.1.11_10-17-2025.zip
    Size: 664693 bytes (649 KiB)
    SHA256: 81018243ED078AA3ACCCFD9E88EA189D4A79C8BA8B0998A6EF44AD041B324A78

    One of the folders has this in it:

    Marvell_AQtion_3.1.11.0_installer_win1x.msi 593,520 bytes

    and right-clicking on that should work just as well as using the .INF setup in
    the other folder.

    *******

    Over file sharing, I got 1.03GB/sec while transferring a 7GB DVD between RAM drives.
    So that's file sharing.

    The servethehome got 9000 on iperf, and dividing that by 8 gives 1.125GB/sec >so my measured value is a little less.

    *******

    The "Random number" of 700Mbit/sec does not ring any bells.

    If you're watching the Ethernet interface in Task Manager, does the
    transfer rate change with time ? My somewhat more modest network
    connection, has a downward glitch every 5 seconds or so, which is
    presumably an issue with the box at the corner. And mine is a wired connection.
    The plan for the 10GbE cards was purely for transfers between machines.

    Paul


    OK, thanks Paul. But I wonder why if I use A flash drive Linux Mint,
    get full speed symmetrical, but on Windows I used to get 1800 Mbit
    down and now reduced to 700 Mbit.
    So which driver should I use?

    --- PyGate Linux v1.5.16
    * Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)
  • From Dan@3:633/10 to All on Monday, June 15, 2026 12:53:07
    On Mon, 15 Jun 2026 03:55:52 -0400, Paul <nospam@needed.invalid>
    wrote:

    On Sun, 6/14/2026 1:02 PM, Dan wrote:
    Hello,

    I am on Community fibre, symmetrical 2.5 gbit. On Nperf I get 1.7 down
    and 2.4 up.
    Upload is 2.4 but download has decreased to 700 Mbit. I am using Ookla
    app in Win 11.
    My Ethernet is a TP link TX 401 10 Gig. Cables, all are Cat 8. Tried
    different ones with the same result. My ONT is 10 Gig capable.
    Got the latest driver from TP link, installed it and no difference.

    Tested with Linux mint and I get 2.4 down and up.


    I have three TX 401, and the testing stopped after I got two installed.
    (I have no 10GbE switch for them, which is why the project is stalled
    waiting for some cheep RealTek based box to "come along".)

    I'm running them on CAT6A. In point to point test, I got around 1000MB/sec >instead of a number derived from 10GbE.

    The purpose of selecting the TX 401, is so the PCI Express slot
    does not have to be too fancy. That's so the card could work in
    the 4930K machine. You need enough lanes (it's an x4 card) and
    enough lane speed, to make it to 10GbE. And none of my equipment
    is exactly smoking-fast. On older equipment, you have to take
    the sub-sized buffers into account in the chipset -- the "lanes
    need to be twice as fast as needed" to cover the buffer issue.
    Later equipment seemed to fix that somehow. PCIe has gotten closer
    to wirespeed since the white paper on buffer size versus speed came out.

    There will be some cheaper ones, from RealTek, which run on
    one PCI Express Rev4 lane, but not many machines will be kitted
    out properly for that particular one. Whereas the TX 401 will run
    in more of your machines (maybe even a 10-12 year old machine).
    A second RealTek chip uses two PCIe Rev3 lanes (and presumably on an x4 shaped >piece of PCB). Which would not be the best for my 4930K with PCIe Rev2
    lanes a-plenty. I think that's where the card is in that machine, on a
    PCIe Rev2 interface. RealTek does not appear destined to make a
    four lane version (their plan, after all, is to make dirt cheap
    kit, even if it sucks around the edges a bit).

    I don't have a 10GbE Switch yet for the thing, which is why I have
    tested them point to point. And my project is in limbo, as I wait
    for worthwhile materials to buy, to finish the job.

    The TX 401 on this machine, is Code 22 (as I have disabled
    it on purpose (to save power while it is not doing anything).
    Driver is Marvel AQtion 3.1.10.0 Tuesday 4/23/2024 . (What catalog.update.microsoft.com
    seems to be serving, is 3.1.10.0 . Tested this, by installing the third card to see
    what driver is handed out, and that is the auto-installed one.)

    The problem with testing them with "file sharing", is the
    link is encrypted, and that could be a limitation for that
    style of testing. As part of your "transport", there should
    be less of a restriction in the path there.

    https://www.servethehome.com/tp-link-10gigabit-pcie-network-adapter-mini-review-marvell-aqc113/

    # It kinda looks like a CSP (chip scale) package, but maybe it's considered to be a fine pitch BGA
    # Only important if you suspected the drivers for the three entries were somehow different.
    # DO NOT remove the heatsink, unless you know you have the thermal tape or paste suited
    # to the chip. I don't know if it is booby trapped, for example.

    Marvell AQC113-B1-C chip

    https://www.marvell.com/support/downloads.html

    Category = Marvell Public Drivers
    Platform = Windows
    Part Number = AQC113
    Keyword = <blank>

    10/17/25 Marvell AQtion Windows 10 and 11 64-bit Driver
    Marvell AQtion Windows 10 and 11 64-bit Driver
    Marvell Public Drivers
    Windows 10 and Windows 11 64-bit driver, windows-10-64-bit-driver Driver, Drivers 3.1.11

    Name: Marvell_AQtion_Win_v3.1.11_10-17-2025.zip
    Size: 664693 bytes (649 KiB)
    SHA256: 81018243ED078AA3ACCCFD9E88EA189D4A79C8BA8B0998A6EF44AD041B324A78

    One of the folders has this in it:

    Marvell_AQtion_3.1.11.0_installer_win1x.msi 593,520 bytes

    and right-clicking on that should work just as well as using the .INF setup in
    the other folder.

    *******

    Over file sharing, I got 1.03GB/sec while transferring a 7GB DVD between RAM drives.
    So that's file sharing.

    The servethehome got 9000 on iperf, and dividing that by 8 gives 1.125GB/sec >so my measured value is a little less.

    *******

    The "Random number" of 700Mbit/sec does not ring any bells.

    If you're watching the Ethernet interface in Task Manager, does the
    transfer rate change with time ? My somewhat more modest network
    connection, has a downward glitch every 5 seconds or so, which is
    presumably an issue with the box at the corner. And mine is a wired connection.
    The plan for the 10GbE cards was purely for transfers between machines.

    Paul



    Update:

    I installed these drivers as instructed by you, sadly no difference.


    Marvell AQC113-B1-C chip

    https://www.marvell.com/support/downloads.html

    Category = Marvell Public Drivers
    Platform = Windows
    Part Number = AQC113
    Keyword = <blank>

    10/17/25 Marvell AQtion Windows 10 and 11 64-bit Driver
    Marvell AQtion Windows 10 and 11 64-bit Driver
    Marvell Public Drivers
    Windows 10 and Windows 11 64-bit driver, windows-10-64-bit-driver Driver, Drivers 3.1.11

    Name: Marvell_AQtion_Win_v3.1.11_10-17-2025.zip
    Size: 664693 bytes (649 KiB)
    SHA256:
    81018243ED078AA3ACCCFD9E88EA189D4A79C8BA8B0998A6EF44AD041B324A78

    One of the folders has this in it:

    Marvell_AQtion_3.1.11.0_installer_win1x.msi 593,520 bytes

    and right-clicking on that should work just as well as using the
    .INF setup in
    the other folder.

    --- PyGate Linux v1.5.16
    * Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)
  • From ....winston@3:633/10 to All on Monday, June 15, 2026 09:46:38
    On 06/15/2026 2:42 AM, Dan wrote:
    On Sun, 14 Jun 2026 17:01:45 -0400, "....winston"
    <winstonmvp@gmail.com> wrote:

    On 06/14/2026 1:02 PM, Dan wrote:
    Hello,

    I am on Community fibre, symmetrical 2.5 gbit. On Nperf I get 1.7 down
    and 2.4 up.
    Upload is 2.4 but download has decreased to 700 Mbit. I am using Ookla
    app in Win 11.
    My Ethernet is a TP link TX 401 10 Gig. Cables, all are Cat 8. Tried
    different ones with the same result. My ONT is 10 Gig capable.
    Got the latest driver from TP link, installed it and no difference.

    Tested with Linux mint and I get 2.4 down and up.

    Router between fiber and device?
    Device using wifi or cable?
    Device speed test performed on wifi or cable
    Device's network card/adapter specs?
    Using a VPN? If so, disable and retest.

    CAT8 is usually overkill, but acceptable.

    For Windows, is TCP auto-tuning enabled?
    Run this command in a Powershell admin:
    netsh interface tcp show global

    What does Auto-tuning show?
    - A 'Normal' result = Enabled

    As a test, if Normal, disable and retest
    netsh interface tcp set global autotuninglevel=disabled
    Restart the device, then retest

    Note Auto-tuning can be turned back on:
    netsh interface tcp set global autotuninglevel=normal





    Microsoft Windows [Version 10.0.26200.8655]
    (c) Microsoft Corporation. All rights reserved.

    C:\Windows\System32>netsh interface tcp show global
    Querying active state...

    TCP Global Parameters
    ----------------------------------------------
    Receive-Side Scaling State : enabled
    Receive Window Auto-Tuning Level : disabled
    Add-On Congestion Control Provider : default
    ECN Capability : disabled
    RFC 1323 Timestamps : allowed
    Initial RTO : 1000
    Receive Segment Coalescing State : enabled
    Non Sack Rtt Resiliency : disabled
    Max SYN Retransmissions : 4
    Fast Open : enabled
    Fast Open Fallback : enabled
    HyStart : enabled
    Proportional Rate Reduction : enabled
    Pacing Profile : off


    C:\Windows\System32>






    Router between fiber and device? Cat 8 Ethernet
    Device using wifi or cable? Cat 8 Ethernet
    Device speed test performed on wifi or cable Cat 8 Ethernet
    Device's network card/adapter specs? TP link TX 401 10 Gig 10 Gig
    enabled
    Using a VPN? If so, disable and retest. No VPN


    Thanks for the info.
    Afiacs, based on your reply, no router present in chain limiting one
    os and not the other.

    Also read your replies to Paul with respect to driver.
    Atm, have no other questions or comments.


    --
    ...wнд?ё?д

    --- PyGate Linux v1.5.16
    * Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)
  • From Paul@3:633/10 to All on Tuesday, June 16, 2026 08:59:58
    On Mon, 6/15/2026 7:53 AM, Dan wrote:
    On Mon, 15 Jun 2026 03:55:52 -0400, Paul <nospam@needed.invalid>
    wrote:

    On Sun, 6/14/2026 1:02 PM, Dan wrote:
    Hello,

    I am on Community fibre, symmetrical 2.5 gbit. On Nperf I get 1.7 down
    and 2.4 up.
    Upload is 2.4 but download has decreased to 700 Mbit. I am using Ookla
    app in Win 11.
    My Ethernet is a TP link TX 401 10 Gig. Cables, all are Cat 8. Tried
    different ones with the same result. My ONT is 10 Gig capable.
    Got the latest driver from TP link, installed it and no difference.

    Tested with Linux mint and I get 2.4 down and up.


    I have three TX 401, and the testing stopped after I got two installed.
    (I have no 10GbE switch for them, which is why the project is stalled
    waiting for some cheep RealTek based box to "come along".)

    I'm running them on CAT6A. In point to point test, I got around 1000MB/sec >> instead of a number derived from 10GbE.

    The purpose of selecting the TX 401, is so the PCI Express slot
    does not have to be too fancy. That's so the card could work in
    the 4930K machine. You need enough lanes (it's an x4 card) and
    enough lane speed, to make it to 10GbE. And none of my equipment
    is exactly smoking-fast. On older equipment, you have to take
    the sub-sized buffers into account in the chipset -- the "lanes
    need to be twice as fast as needed" to cover the buffer issue.
    Later equipment seemed to fix that somehow. PCIe has gotten closer
    to wirespeed since the white paper on buffer size versus speed came out.

    There will be some cheaper ones, from RealTek, which run on
    one PCI Express Rev4 lane, but not many machines will be kitted
    out properly for that particular one. Whereas the TX 401 will run
    in more of your machines (maybe even a 10-12 year old machine).
    A second RealTek chip uses two PCIe Rev3 lanes (and presumably on an x4 shaped
    piece of PCB). Which would not be the best for my 4930K with PCIe Rev2
    lanes a-plenty. I think that's where the card is in that machine, on a
    PCIe Rev2 interface. RealTek does not appear destined to make a
    four lane version (their plan, after all, is to make dirt cheap
    kit, even if it sucks around the edges a bit).

    I don't have a 10GbE Switch yet for the thing, which is why I have
    tested them point to point. And my project is in limbo, as I wait
    for worthwhile materials to buy, to finish the job.

    The TX 401 on this machine, is Code 22 (as I have disabled
    it on purpose (to save power while it is not doing anything).
    Driver is Marvel AQtion 3.1.10.0 Tuesday 4/23/2024 . (What catalog.update.microsoft.com
    seems to be serving, is 3.1.10.0 . Tested this, by installing the third card to see
    what driver is handed out, and that is the auto-installed one.)

    The problem with testing them with "file sharing", is the
    link is encrypted, and that could be a limitation for that
    style of testing. As part of your "transport", there should
    be less of a restriction in the path there.

    https://www.servethehome.com/tp-link-10gigabit-pcie-network-adapter-mini-review-marvell-aqc113/

    # It kinda looks like a CSP (chip scale) package, but maybe it's considered to be a fine pitch BGA
    # Only important if you suspected the drivers for the three entries were somehow different.
    # DO NOT remove the heatsink, unless you know you have the thermal tape or paste suited
    # to the chip. I don't know if it is booby trapped, for example.

    Marvell AQC113-B1-C chip

    https://www.marvell.com/support/downloads.html

    Category = Marvell Public Drivers
    Platform = Windows
    Part Number = AQC113
    Keyword = <blank>

    10/17/25 Marvell AQtion Windows 10 and 11 64-bit Driver
    Marvell AQtion Windows 10 and 11 64-bit Driver
    Marvell Public Drivers
    Windows 10 and Windows 11 64-bit driver, windows-10-64-bit-driver Driver, Drivers 3.1.11

    Name: Marvell_AQtion_Win_v3.1.11_10-17-2025.zip
    Size: 664693 bytes (649 KiB)
    SHA256: 81018243ED078AA3ACCCFD9E88EA189D4A79C8BA8B0998A6EF44AD041B324A78 >>
    One of the folders has this in it:

    Marvell_AQtion_3.1.11.0_installer_win1x.msi 593,520 bytes

    and right-clicking on that should work just as well as using the .INF setup in
    the other folder.

    *******

    Over file sharing, I got 1.03GB/sec while transferring a 7GB DVD between RAM drives.
    So that's file sharing.

    The servethehome got 9000 on iperf, and dividing that by 8 gives 1.125GB/sec >> so my measured value is a little less.

    *******

    The "Random number" of 700Mbit/sec does not ring any bells.

    If you're watching the Ethernet interface in Task Manager, does the
    transfer rate change with time ? My somewhat more modest network
    connection, has a downward glitch every 5 seconds or so, which is
    presumably an issue with the box at the corner. And mine is a wired connection.
    The plan for the 10GbE cards was purely for transfers between machines.

    Paul



    Update:

    I installed these drivers as instructed by you, sadly no difference.


    Marvell AQC113-B1-C chip

    https://www.marvell.com/support/downloads.html

    Category = Marvell Public Drivers
    Platform = Windows
    Part Number = AQC113
    Keyword = <blank>

    10/17/25 Marvell AQtion Windows 10 and 11 64-bit Driver
    Marvell AQtion Windows 10 and 11 64-bit Driver
    Marvell Public Drivers
    Windows 10 and Windows 11 64-bit driver, windows-10-64-bit-driver Driver, Drivers 3.1.11

    Name: Marvell_AQtion_Win_v3.1.11_10-17-2025.zip
    Size: 664693 bytes (649 KiB)
    SHA256:
    81018243ED078AA3ACCCFD9E88EA189D4A79C8BA8B0998A6EF44AD041B324A78

    One of the folders has this in it:

    Marvell_AQtion_3.1.11.0_installer_win1x.msi 593,520 bytes

    and right-clicking on that should work just as well as using the
    .INF setup in
    the other folder.


    I was not expecting that to make a difference. That was a long shot.
    The chip has been around for a while, so the datapath behavior should be stable.

    We would need to understand the NDIS feature set in the
    control panel, and see if anything at the other end of the
    connection is somehow "set differently". It's possible
    the Linux settings for their control panel are different
    than the AQC113 control panel on Windows.

    It should be possible to capture that interface in Wireshark, on the
    PC end. But I don't know if an attempt at an interpretation
    of the trace, would be all that easy. For example, if the screen
    was thick with black packets, sometimes eth links get into
    retries ("duplicate packet"). And again, you may not be able to understand exactly why that is happening (sequence number issue ? CRC errors ?).

    I don't think I have the skill to debug this.

    Even if you collected a wireshark trace in Linux and a wireshark trace
    in Windows, you may not be able to figure out which of those "options"
    in the windows control panel, is doing it. I don't expect, if we
    compared our settings, we would find any differences on the default
    values. And then, why would the far end vary its choices for link
    startup parameters, just because a Windows or a Linux box was on the
    other end ?

    Paul

    --- PyGate Linux v1.5.17
    * Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)
  • From Java Jive@3:633/10 to All on Tuesday, June 16, 2026 15:29:28
    On 2026-06-15 12:37, Dan wrote:

    OK, thanks Paul. But I wonder why if I use A flash drive Linux Mint,
    get full speed symmetrical, but on Windows I used to get 1800 Mbit
    down and now reduced to 700 Mbit.
    So which driver should I use?

    Although now officially retired, I still have one or two legacy
    'customers' who call me in on an ad hoc basis to sort out problems for
    them. Reading through the thread, my first instinct was that I couldn't
    help, but now I'm beginning to wonder if an experience I had at one of
    these 'customers' may be relevant.

    It was a new FTTP install, after everything had been set up and running
    I tried a speed test on my test laptop, an old Dell Latitude D610 which
    has a Gigabit netcard, and got well below the supposed speed of the connection. Alarmed, after some thought, I decided to test again with a different machine, in case that old laptop was throwing a wobbly I
    hadn't noticed before, so I tested again with a Dell Inspiron 15RSE
    7520, and that successfully maxed out the connection. The problem was
    simply that the older laptop, despite its Gigabit hardware, hadn't got
    the grunt to sustain a level of network activity sufficient to max out
    the connection.

    In your case, something similar seems to be happening only this time on
    the same laptop. The successful Linux test shows that the connection is running at the appropriate speed, so, as everyone has immediately
    realised but I'll restate the obvious, the problem is clearly something
    in Windows. Try looking/thinking back to when the reduction in speed occurred, what was going on at the time that could have changed things?
    Likely candidates are malware or legitimate software such as two
    conflicting anti-virus programs crippling the machine so the hardware is
    too busy to max out the network connection. What updates were installed
    at the time? Did you install any new software at the time? Did you
    encrypt the machine then? Etc.

    --

    Fake news kills!

    I may be contacted via the contact address given on my website: www.macfh.co.uk


    --- PyGate Linux v1.5.17
    * Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)
  • From Peter Johnson@3:633/10 to All on Tuesday, June 16, 2026 18:13:47
    On Sun, 14 Jun 2026 18:02:50 +0100, Dan
    <danZZZ12398newsgroups@outlook.com> wrote:

    Hello,

    I am on Community fibre, symmetrical 2.5 gbit. On Nperf I get 1.7 down
    and 2.4 up.
    Upload is 2.4 but download has decreased to 700 Mbit. I am using Ookla
    app in Win 11.
    My Ethernet is a TP link TX 401 10 Gig. Cables, all are Cat 8. Tried >different ones with the same result. My ONT is 10 Gig capable.
    Got the latest driver from TP link, installed it and no difference.

    Tested with Linux mint and I get 2.4 down and up.

    My City Fibre 2.5Gb connection wasn't normal this morning (16th). The
    ping was much higher than usual and the download/upload asymetric
    instead of symetric. I don't check it very often but something I was
    browsing wasn't loading properly.
    Just checked it again, having read your posting, and the
    download/upload are back to normal and the ping is 7. Usually ping is
    4 but I'm not going to lose any sleep over that.
    If there had been no improvement I'd have got on to the supplier,
    which is a move that I'd recommend if you don't see any improvement. .

    --- PyGate Linux v1.5.17
    * Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)
  • From Paul@3:633/10 to All on Tuesday, June 16, 2026 18:03:58
    On Tue, 6/16/2026 10:29 AM, Java Jive wrote:
    On 2026-06-15 12:37, Dan wrote:

    OK, thanks Paul. But I wonder why if I use A flash drive Linux Mint,
    get full speed symmetrical, but on Windows I used to get 1800 Mbit
    down and now reduced to 700 Mbit.
    So which driver should I use?

    Although now officially retired, I still have one or two legacy 'customers' who call me in on an ad hoc basis to sort out problems for them.  Reading through the thread, my first instinct was that I couldn't help, but now I'm beginning to wonder if an experience I had at one of these 'customers' may be relevant.

    It was a new FTTP install, after everything had been set up and running I tried a speed test on my test laptop, an old Dell Latitude D610 which has a Gigabit netcard, and got well below the supposed speed of the connection.  Alarmed, after some thought, I decided to test again with a different machine, in case that old laptop was throwing a wobbly I hadn't noticed before, so I tested again with a Dell Inspiron 15RSE 7520, and that successfully maxed out the connection.  The problem was simply that the older laptop, despite its Gigabit hardware, hadn't got the grunt to sustain a level of network activity sufficient to max out the connection.

    In your case, something similar seems to be happening only this time on the same laptop.  The successful Linux test shows that the connection is running at the appropriate speed, so, as everyone has immediately realised but I'll restate the obvious, the problem is clearly something in Windows.  Try looking/thinking back to when the reduction in speed occurred, what was going on at the time that could have changed things? Likely candidates are malware or legitimate software such as two conflicting anti-virus programs crippling the machine so the hardware is too busy to max out the network connection.  What updates were installed at the time?  Did you install any new software at the time?  Did you encrypt the machine then?  Etc.


    There was a RealTek GbE which emits *5* interrupts per packet.
    This has a detrimental effect on performance, and after some
    testing at the time, I determined an overclock to 4GHz would be
    required to make that device work properly. Whereas a Marvell GbE
    could do the 112MB/sec speed of GbE file sharing, and still have
    CPU left over. The RealTek GbE could do 70-80MB/sec because
    it was just showering the CPU in interrupts.

    The AQC113 has Adaptive Interrupt Moderation, which means
    there is one interrupt in Ring 0, a quick ACK, a DPC is queued,
    and multiple packet statuses are processed in the Ring 3 DPC
    processing call. And this "cures" that RealTek problem, and is
    more fancy than that old RealTek part.

    The reason the RealTek had five interrupts, is there was an
    interrupt per portion-of-packet transferred. You could store packet
    headers one place in the driver, packet bodies somewhere else,
    maybe with the intention of a Zero Copy architecture. Maybe the massive
    number of Interrupts, meant not paying a licensing fee for Interrupt
    Moderation patent ?

    I would not use Jumbo packets. Maybe a Jumbo packet choice would work
    in a direct-to-ISP situation, but I don't know anything about their
    settings choices in such a connection. My "broadband" has always
    been pretty slow (not the broad kind of broadband). I've tested Jumbo
    in a home LAN environment, and concluded with the limitations, it
    wasn't worth the aggravation. Stick with the same rough 1500 byte MTU
    numbers everywhere.

    When you don't use a router in the path, and the incoming protocol is
    PPPOE or PPPOA (point to point protocol over ATM), in the latter
    case the incoming packets would be 53 bytes and that's a hell of
    a lot of segmentation and reassembly to reach 2.4Gbit/sec made out
    of 53 byte chunks. When you have a router, it removes whatever the
    protocol is on the link, and delivers more standard Ethernet packets
    to the household (from the router).

    When I first got the Internetz at home, I was running without
    a router, and the CPU was handling incoming PPPOE and making Eth
    from it. And it did "work better" when I later got a router
    and it handled the PPPOE packets.

    To do that, you need a router that "has the poop to do
    2.4Gbit/sec processing". There are going to be plenty of
    routers not up to the task. It requires considerable Googling
    using the name-of-your-service, to find other adventurers
    who have bought a third party router, to find one powerful
    enough to run the contracted wirespeed rate.

    And you could try Wireshark on that single-computer-to-ONT you've
    set up, capture a sampling and try and understand the characteristics
    of what is being used. See whether there are 53 byte packets.
    Check whether a CPU core rails during a transfer. You know,
    pretend to be a scientist or something :-) Wear a lab coat.
    The way your ISP wanted it to be for you :-) When I've looked
    at articles in the past on DSLReports and the hell some people
    go through getting a high speed connection to "run at the
    contracted rate". It really is rocket science.

    Paul

    --- PyGate Linux v1.5.17
    * Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)
  • From Brian Gregory@3:633/10 to All on Tuesday, June 16, 2026 23:48:33
    Check also that Ethernet flow control is disabled in your NIC.

    --
    Brian Gregory (in England).

    --- PyGate Linux v1.5.17
    * Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)
  • From Dan@3:633/10 to All on Wednesday, June 17, 2026 07:04:24
    On Tue, 16 Jun 2026 18:13:47 +0100, Peter Johnson
    <peter@parksidewood.nospam> wrote:

    On Sun, 14 Jun 2026 18:02:50 +0100, Dan
    <danZZZ12398newsgroups@outlook.com> wrote:

    Hello,

    I am on Community fibre, symmetrical 2.5 gbit. On Nperf I get 1.7 down
    and 2.4 up.
    Upload is 2.4 but download has decreased to 700 Mbit. I am using Ookla
    app in Win 11.
    My Ethernet is a TP link TX 401 10 Gig. Cables, all are Cat 8. Tried >>different ones with the same result. My ONT is 10 Gig capable.
    Got the latest driver from TP link, installed it and no difference.

    Tested with Linux mint and I get 2.4 down and up.

    My City Fibre 2.5Gb connection wasn't normal this morning (16th). The
    ping was much higher than usual and the download/upload asymetric
    instead of symetric. I don't check it very often but something I was
    browsing wasn't loading properly.
    Just checked it again, having read your posting, and the
    download/upload are back to normal and the ping is 7. Usually ping is
    4 but I'm not going to lose any sleep over that.
    If there had been no improvement I'd have got on to the supplier,
    which is a move that I'd recommend if you don't see any improvement. .


    Thanks to all. In Linux Mint I get 2.4 symmetrical. So it is not a
    hardware problem.
    In nPerf, I get 1500 down and 2.4 up.
    CPU is a AMD 7 9700X in the test machine.
    Ping is about 3.
    Anti Virus software is Norton. Been using that for years with no
    connection problems.
    No boot SSD encryption.
    "Check also that Ethernet flow control is disabled in your NIC."
    How do I do that?
    Wireshark is way too technical for me.

    --- PyGate Linux v1.5.17
    * Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)
  • From Paul@3:633/10 to All on Wednesday, June 17, 2026 05:20:21
    On Wed, 6/17/2026 2:04 AM, Dan wrote:
    On Tue, 16 Jun 2026 18:13:47 +0100, Peter Johnson
    <peter@parksidewood.nospam> wrote:

    On Sun, 14 Jun 2026 18:02:50 +0100, Dan
    <danZZZ12398newsgroups@outlook.com> wrote:

    Hello,

    I am on Community fibre, symmetrical 2.5 gbit. On Nperf I get 1.7 down
    and 2.4 up.
    Upload is 2.4 but download has decreased to 700 Mbit. I am using Ookla
    app in Win 11.
    My Ethernet is a TP link TX 401 10 Gig. Cables, all are Cat 8. Tried
    different ones with the same result. My ONT is 10 Gig capable.
    Got the latest driver from TP link, installed it and no difference.

    Tested with Linux mint and I get 2.4 down and up.

    My City Fibre 2.5Gb connection wasn't normal this morning (16th). The
    ping was much higher than usual and the download/upload asymetric
    instead of symetric. I don't check it very often but something I was
    browsing wasn't loading properly.
    Just checked it again, having read your posting, and the
    download/upload are back to normal and the ping is 7. Usually ping is
    4 but I'm not going to lose any sleep over that.
    If there had been no improvement I'd have got on to the supplier,
    which is a move that I'd recommend if you don't see any improvement. .


    Thanks to all. In Linux Mint I get 2.4 symmetrical. So it is not a
    hardware problem.
    In nPerf, I get 1500 down and 2.4 up.
    CPU is a AMD 7 9700X in the test machine.
    Ping is about 3.
    Anti Virus software is Norton. Been using that for years with no
    connection problems.
    No boot SSD encryption.
    "Check also that Ethernet flow control is disabled in your NIC."
    How do I do that?
    Wireshark is way too technical for me.


    Right-click Start, look for Device Manager.

    In Device Manager, find the TX 401 as "Marvel AQtion 10Gbit Network Adapter". Double click.

    Select the Advanced tab.

    For Flow Control, there are these options:

    Disabled
    RX Enabled
    TX Enabled
    RX & TX Enabled <=== default

    Paul


    --- PyGate Linux v1.5.17
    * Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)
  • From Dan@3:633/10 to All on Wednesday, June 17, 2026 12:12:16
    On Wed, 17 Jun 2026 05:20:21 -0400, Paul <nospam@needed.invalid>
    wrote:

    On Wed, 6/17/2026 2:04 AM, Dan wrote:
    On Tue, 16 Jun 2026 18:13:47 +0100, Peter Johnson
    <peter@parksidewood.nospam> wrote:

    On Sun, 14 Jun 2026 18:02:50 +0100, Dan
    <danZZZ12398newsgroups@outlook.com> wrote:

    Hello,

    I am on Community fibre, symmetrical 2.5 gbit. On Nperf I get 1.7 down >>>> and 2.4 up.
    Upload is 2.4 but download has decreased to 700 Mbit. I am using Ookla >>>> app in Win 11.
    My Ethernet is a TP link TX 401 10 Gig. Cables, all are Cat 8. Tried
    different ones with the same result. My ONT is 10 Gig capable.
    Got the latest driver from TP link, installed it and no difference.

    Tested with Linux mint and I get 2.4 down and up.

    My City Fibre 2.5Gb connection wasn't normal this morning (16th). The
    ping was much higher than usual and the download/upload asymetric
    instead of symetric. I don't check it very often but something I was
    browsing wasn't loading properly.
    Just checked it again, having read your posting, and the
    download/upload are back to normal and the ping is 7. Usually ping is
    4 but I'm not going to lose any sleep over that.
    If there had been no improvement I'd have got on to the supplier,
    which is a move that I'd recommend if you don't see any improvement. .


    Thanks to all. In Linux Mint I get 2.4 symmetrical. So it is not a
    hardware problem.
    In nPerf, I get 1500 down and 2.4 up.
    CPU is a AMD 7 9700X in the test machine.
    Ping is about 3.
    Anti Virus software is Norton. Been using that for years with no
    connection problems.
    No boot SSD encryption.
    "Check also that Ethernet flow control is disabled in your NIC."
    How do I do that?
    Wireshark is way too technical for me.


    Right-click Start, look for Device Manager.

    In Device Manager, find the TX 401 as "Marvel AQtion 10Gbit Network Adapter". >Double click.

    Select the Advanced tab.

    For Flow Control, there are these options:

    Disabled
    RX Enabled
    TX Enabled
    RX & TX Enabled <=== default

    Paul



    Check also that Ethernet flow control is disabled in your NIC


    In Device Manager, find the TX 401 as "Marvel AQtion 10Gbit Network
    Adapter".
    Double click.

    Select the Advanced tab.

    For Flow Control, there are these options:

    Disabled
    RX Enabled
    TX Enabled
    RX & TX Enabled <=== default


    I set it to disabled, sadly no difference.

    --- PyGate Linux v1.5.17
    * Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)
  • From Paul@3:633/10 to All on Wednesday, June 17, 2026 07:36:11
    On Wed, 6/17/2026 7:12 AM, Dan wrote:
    On Wed, 17 Jun 2026 05:20:21 -0400, Paul <nospam@needed.invalid>
    wrote:

    On Wed, 6/17/2026 2:04 AM, Dan wrote:
    On Tue, 16 Jun 2026 18:13:47 +0100, Peter Johnson
    <peter@parksidewood.nospam> wrote:

    On Sun, 14 Jun 2026 18:02:50 +0100, Dan
    <danZZZ12398newsgroups@outlook.com> wrote:

    Hello,

    I am on Community fibre, symmetrical 2.5 gbit. On Nperf I get 1.7 down >>>>> and 2.4 up.
    Upload is 2.4 but download has decreased to 700 Mbit. I am using Ookla >>>>> app in Win 11.
    My Ethernet is a TP link TX 401 10 Gig. Cables, all are Cat 8. Tried >>>>> different ones with the same result. My ONT is 10 Gig capable.
    Got the latest driver from TP link, installed it and no difference.

    Tested with Linux mint and I get 2.4 down and up.

    My City Fibre 2.5Gb connection wasn't normal this morning (16th). The
    ping was much higher than usual and the download/upload asymetric
    instead of symetric. I don't check it very often but something I was
    browsing wasn't loading properly.
    Just checked it again, having read your posting, and the
    download/upload are back to normal and the ping is 7. Usually ping is
    4 but I'm not going to lose any sleep over that.
    If there had been no improvement I'd have got on to the supplier,
    which is a move that I'd recommend if you don't see any improvement. .


    Thanks to all. In Linux Mint I get 2.4 symmetrical. So it is not a
    hardware problem.
    In nPerf, I get 1500 down and 2.4 up.
    CPU is a AMD 7 9700X in the test machine.
    Ping is about 3.
    Anti Virus software is Norton. Been using that for years with no
    connection problems.
    No boot SSD encryption.
    "Check also that Ethernet flow control is disabled in your NIC."
    How do I do that?
    Wireshark is way too technical for me.


    Right-click Start, look for Device Manager.

    In Device Manager, find the TX 401 as "Marvel AQtion 10Gbit Network Adapter".
    Double click.

    Select the Advanced tab.

    For Flow Control, there are these options:

    Disabled
    RX Enabled
    TX Enabled
    RX & TX Enabled <=== default

    Paul



    Check also that Ethernet flow control is disabled in your NIC


    In Device Manager, find the TX 401 as "Marvel AQtion 10Gbit Network
    Adapter".
    Double click.

    Select the Advanced tab.

    For Flow Control, there are these options:

    Disabled
    RX Enabled
    TX Enabled
    RX & TX Enabled <=== default


    I set it to disabled, sadly no difference.


    This might well be an ISP problem, and you should contact support
    and see if they have any status items that could be affecting
    your speed. Yes, I know one OS is able to run the equipment
    properly, but we're attempting to tap Support for some clues
    or ideas.

    A technique I use occasionally, is to do a clean Windows install, then
    carry out the speed test again, to see if Windows can be coaxed
    to work. Then at least, you would know that something is screwed
    up on your daily driver OS install. Since you're running a single
    PC directly off the interface, and not using a router, then the
    format on the cable (like PPPOE) must be de-capsulated by your OS.
    My point-to-point testing would just use ordinary Ethernet packets
    and PPPOE would not be involved. The router, if present, would
    remove PPPOE and convert the incoming WAN signal to plain Ethernet.

    Wireshark is available for both Windows and Linux, and all I would
    be looking for there, is "black packets". Sometimes these indicate
    "duplicate packet", as an example of a flaw you sometimes see. There
    is no particular reason to examine any minutiae with the Wireshark.
    it would just be for a quick check for black packets.

    I've been lucky so far, that when I bench my cards in point
    to point, the transfer speed is in the right ball park. My result isn't
    exactly the same as the "servethehome" review result (iPerf at 9000Mbit/sec) but it isn't off by too much.

    the card may have domain validation, and if there are too many
    CRC errors, maybe it gears down to 1GbE rate instead, or maybe
    5GbE, 2GbE, 1GbE, but it is hard to imagine such an activity causing 700Mbit/sec as the final rate. That does not make a lot of sense.

    Downshift Retries: 4

    I don't even know whether I'd want to play with that.

    One thing that is interesting about that pane, is I don't
    see a forcing function, to "make" the interface run at 1GbE
    instead of 10GbE. Normally on NICs, you would expect to see
    "Auto, 1GbE, 100Mbit/sec, 10Mbit/sec" and one of the Marvell
    chips goes all the way down to 10Mbit/sec. If you set to one
    of the numeric values, that stops attempts at negotiating.
    It would appear then, that the TX 401 on Windows runs "Auto"
    and it will negotiate the highest rate the other end has to offer.
    And only if some "Downshift" criterion is met, might it go from
    10GbE to 5GbE or... lower.

    Paul




    --- PyGate Linux v1.5.17
    * Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)
  • From Dan@3:633/10 to All on Wednesday, June 17, 2026 13:06:39
    On Wed, 17 Jun 2026 07:36:11 -0400, Paul <nospam@needed.invalid>
    wrote:

    On Wed, 6/17/2026 7:12 AM, Dan wrote:
    On Wed, 17 Jun 2026 05:20:21 -0400, Paul <nospam@needed.invalid>
    wrote:

    On Wed, 6/17/2026 2:04 AM, Dan wrote:
    On Tue, 16 Jun 2026 18:13:47 +0100, Peter Johnson
    <peter@parksidewood.nospam> wrote:

    On Sun, 14 Jun 2026 18:02:50 +0100, Dan
    <danZZZ12398newsgroups@outlook.com> wrote:

    Hello,

    I am on Community fibre, symmetrical 2.5 gbit. On Nperf I get 1.7 down >>>>>> and 2.4 up.
    Upload is 2.4 but download has decreased to 700 Mbit. I am using Ookla >>>>>> app in Win 11.
    My Ethernet is a TP link TX 401 10 Gig. Cables, all are Cat 8. Tried >>>>>> different ones with the same result. My ONT is 10 Gig capable.
    Got the latest driver from TP link, installed it and no difference. >>>>>>
    Tested with Linux mint and I get 2.4 down and up.

    My City Fibre 2.5Gb connection wasn't normal this morning (16th). The >>>>> ping was much higher than usual and the download/upload asymetric
    instead of symetric. I don't check it very often but something I was >>>>> browsing wasn't loading properly.
    Just checked it again, having read your posting, and the
    download/upload are back to normal and the ping is 7. Usually ping is >>>>> 4 but I'm not going to lose any sleep over that.
    If there had been no improvement I'd have got on to the supplier,
    which is a move that I'd recommend if you don't see any improvement. . >>>>

    Thanks to all. In Linux Mint I get 2.4 symmetrical. So it is not a
    hardware problem.
    In nPerf, I get 1500 down and 2.4 up.
    CPU is a AMD 7 9700X in the test machine.
    Ping is about 3.
    Anti Virus software is Norton. Been using that for years with no
    connection problems.
    No boot SSD encryption.
    "Check also that Ethernet flow control is disabled in your NIC."
    How do I do that?
    Wireshark is way too technical for me.


    Right-click Start, look for Device Manager.

    In Device Manager, find the TX 401 as "Marvel AQtion 10Gbit Network Adapter".
    Double click.

    Select the Advanced tab.

    For Flow Control, there are these options:

    Disabled
    RX Enabled
    TX Enabled
    RX & TX Enabled <=== default

    Paul



    Check also that Ethernet flow control is disabled in your NIC


    In Device Manager, find the TX 401 as "Marvel AQtion 10Gbit Network
    Adapter".
    Double click.

    Select the Advanced tab.

    For Flow Control, there are these options:

    Disabled
    RX Enabled
    TX Enabled
    RX & TX Enabled <=== default


    I set it to disabled, sadly no difference.


    This might well be an ISP problem, and you should contact support
    and see if they have any status items that could be affecting
    your speed. Yes, I know one OS is able to run the equipment
    properly, but we're attempting to tap Support for some clues
    or ideas.

    A technique I use occasionally, is to do a clean Windows install, then
    carry out the speed test again, to see if Windows can be coaxed
    to work. Then at least, you would know that something is screwed
    up on your daily driver OS install. Since you're running a single
    PC directly off the interface, and not using a router, then the
    format on the cable (like PPPOE) must be de-capsulated by your OS.
    My point-to-point testing would just use ordinary Ethernet packets
    and PPPOE would not be involved. The router, if present, would
    remove PPPOE and convert the incoming WAN signal to plain Ethernet.

    Wireshark is available for both Windows and Linux, and all I would
    be looking for there, is "black packets". Sometimes these indicate
    "duplicate packet", as an example of a flaw you sometimes see. There
    is no particular reason to examine any minutiae with the Wireshark.
    it would just be for a quick check for black packets.

    I've been lucky so far, that when I bench my cards in point
    to point, the transfer speed is in the right ball park. My result isn't >exactly the same as the "servethehome" review result (iPerf at 9000Mbit/sec) >but it isn't off by too much.

    the card may have domain validation, and if there are too many
    CRC errors, maybe it gears down to 1GbE rate instead, or maybe
    5GbE, 2GbE, 1GbE, but it is hard to imagine such an activity causing >700Mbit/sec as the final rate. That does not make a lot of sense.

    Downshift Retries: 4

    I don't even know whether I'd want to play with that.

    One thing that is interesting about that pane, is I don't
    see a forcing function, to "make" the interface run at 1GbE
    instead of 10GbE. Normally on NICs, you would expect to see
    "Auto, 1GbE, 100Mbit/sec, 10Mbit/sec" and one of the Marvell
    chips goes all the way down to 10Mbit/sec. If you set to one
    of the numeric values, that stops attempts at negotiating.
    It would appear then, that the TX 401 on Windows runs "Auto"
    and it will negotiate the highest rate the other end has to offer.
    And only if some "Downshift" criterion is met, might it go from
    10GbE to 5GbE or... lower.

    Paul




    So, I can only fix this by a full re-install of Win 11?

    --- PyGate Linux v1.5.17
    * Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)
  • From Paul@3:633/10 to All on Wednesday, June 17, 2026 09:19:40
    On Wed, 6/17/2026 8:06 AM, Dan wrote:

    So, I can only fix this by a full re-install of Win 11?


    That's a debug technique.

    It does not guarantee a damn thing of course.

    But we would like to know, in a general way, "which end" of the
    setup is at fault.

    You yourself have used differential debug technique, by
    comparing your Linux (works) result versus your Windows (slow) result.
    This is the same technique, only we're comparing a clean Windows install without Avast or AVG installed, versus a Windows install that
    has been around for a while and might have picked up bad habits.

    One thing you have to remember, is not a lot of people connect
    a computer directly to a 2.4Gbit/sec broadband service :-) Your
    configuration should definitely work, and the OS has had the
    capability of doing that termination for a lot of years. But if I were
    to take a survey of users, maybe 0.5% of them might be attempting
    such a thing. With some ISPs, the router is a rental item that
    comes with the service for example, so any PPPOE processing
    needed, just happens to be done by the router (a router so powerful,
    it can handle such a data-rate). Not many routers can decapsulate
    a 2.4Gbit/sec signal at wire speed. My router runs at 1/100th the necessary speed (a wimp of a router), but then no 10GbE signal is going through it.
    You can definitely do this with Windows and the system processor.
    Open Task Manager while you run your Speed Test, and see what kind
    of core-railing is going on.

    *******

    In this example, a very weird symptom. When he appears to be using some Southbridge (PCH) lanes, the card seems to be resetting itself.

    https://www.reddit.com/r/pchelp/comments/1ko2xk3/tplink_tx401_constant_connecting_and_disconnecting/

    One commenter in a thread seems to think the chip can't drive a
    long cable, and there may be a length limit issue. My cable is CAT6A
    and is ten feet long, for whatever that is worth.

    On my machines, one machine is using the upper slot for the TX-401,
    the other machine is using the lower x4 slot for the TX-401. Makes
    no difference, here. No problems being cause by a slow choice.

    There is a Powershell

    https://stackoverflow.com/questions/66279388/any-windows-alternative-to-unix-utility-ethtool-for-autonegotiation-of-ethernet

    Get-NetAdapter

    and friends, a bit limited but something to try. But all the second
    command is doing, is displaying what's in that Device Manager panel.
    There still isn't as much info as I would like.

    While I could look in Eventvwr.msc, I don't know where exactly
    any errors from a NIC would end up.

    *******

    OK, there are some comments here that seem to match yours. Some
    people seem to be moderately happy. Linux always gets good press.
    Windows, not so much.

    https://www.pbtech.com/pacific/product/NICTPL1001/TP-Link-TX401-10Gbps-PCI-E-Ethernet-Adapter-1-x-RJ

    This is the Device Manager "Details" tab for mine, Hardware Ids.
    See if you have the same hardware as mine.

    PCI\VEN_1D6A&DEV_04C0&SUBSYS_00011D6A&REV_03 PCI\VEN_1D6A&DEV_04C0&SUBSYS_00011D6A
    PCI\VEN_1D6A&DEV_04C0&CC_020000
    PCIIVEN_1D6A&DEV_04C0&CC_0200

    Paul



    --- PyGate Linux v1.5.17
    * Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)
  • From Java Jive@3:633/10 to All on Wednesday, June 17, 2026 15:27:37
    On 2026-06-17 13:06, Dan wrote:

    So, I can only fix this by a full re-install of Win 11?

    To amplify, I hope, the reply that Paul has already made ...

    He's not expecting you to throw away your current installation of
    Windows, but rather to back it up with an imaging program such as
    Macrium Reflect, the last free version of which is 8.0.7783, available
    here ...

    For completeness 32-bit version ...

    <https://download.macrium.com/reflect/v8/v8.0.7783/reflect_setup_free_x86.exe>

    ... but more probably what you want, 64-bit version ...

    <https://download.macrium.com/reflect/v8/v8.0.7783/reflect_setup_free_x64.exe>

    To do this you will need a spare hard disk of sufficient capacity to
    hold a back-up disk image, and a means of connecting it during the backup.

    So the procedure would be ...

    1) Backup current install of W11 to some form of external storage;
    2) Reinstall W11;
    3) Run the same speed test again and note the results;
    4) Depending on the results of Step 3, one of:
    Restore the original W11;
    Remake the original W11 by reinstalling in stages everything
    that was on it, testing the speed after each major installation,
    finally copying your data from the original installation.

    Of course, the above is tedious, but it may be the only way to get to
    the bottom of what's causing the problem.

    --

    Fake news kills!

    I may be contacted via the contact address given on my website: www.macfh.co.uk


    --- PyGate Linux v1.5.17
    * Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)
  • From Dan@3:633/10 to All on Wednesday, June 17, 2026 18:14:26
    On Wed, 17 Jun 2026 15:27:37 +0100, Java Jive <java@evij.com.invalid>
    wrote:

    On 2026-06-17 13:06, Dan wrote:

    So, I can only fix this by a full re-install of Win 11?

    To amplify, I hope, the reply that Paul has already made ...

    He's not expecting you to throw away your current installation of
    Windows, but rather to back it up with an imaging program such as
    Macrium Reflect, the last free version of which is 8.0.7783, available
    here ...

    For completeness 32-bit version ...

    <https://download.macrium.com/reflect/v8/v8.0.7783/reflect_setup_free_x86.exe>

    ... but more probably what you want, 64-bit version ...

    <https://download.macrium.com/reflect/v8/v8.0.7783/reflect_setup_free_x64.exe>

    To do this you will need a spare hard disk of sufficient capacity to
    hold a back-up disk image, and a means of connecting it during the backup.

    So the procedure would be ...

    1) Backup current install of W11 to some form of external storage;
    2) Reinstall W11;
    3) Run the same speed test again and note the results;
    4) Depending on the results of Step 3, one of:
    Restore the original W11;
    Remake the original W11 by reinstalling in stages everything
    that was on it, testing the speed after each major installation,
    finally copying your data from the original installation.

    Of course, the above is tedious, but it may be the only way to get to
    the bottom of what's causing the problem.


    Thank you all. I will have to bak up my data, image and then
    re-install..

    So the x64 version of macrium works wih Win 11 25H2?

    --- PyGate Linux v1.5.17
    * Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)
  • From Dan@3:633/10 to All on Wednesday, June 17, 2026 18:18:56
    On Wed, 17 Jun 2026 09:19:40 -0400, Paul <nospam@needed.invalid>
    wrote:

    On Wed, 6/17/2026 8:06 AM, Dan wrote:

    So, I can only fix this by a full re-install of Win 11?


    That's a debug technique.

    It does not guarantee a damn thing of course.

    But we would like to know, in a general way, "which end" of the
    setup is at fault.

    You yourself have used differential debug technique, by
    comparing your Linux (works) result versus your Windows (slow) result.
    This is the same technique, only we're comparing a clean Windows install >without Avast or AVG installed, versus a Windows install that
    has been around for a while and might have picked up bad habits.

    One thing you have to remember, is not a lot of people connect
    a computer directly to a 2.4Gbit/sec broadband service :-) Your
    configuration should definitely work, and the OS has had the
    capability of doing that termination for a lot of years. But if I were
    to take a survey of users, maybe 0.5% of them might be attempting
    such a thing. With some ISPs, the router is a rental item that
    comes with the service for example, so any PPPOE processing
    needed, just happens to be done by the router (a router so powerful,
    it can handle such a data-rate). Not many routers can decapsulate
    a 2.4Gbit/sec signal at wire speed. My router runs at 1/100th the necessary >speed (a wimp of a router), but then no 10GbE signal is going through it.
    You can definitely do this with Windows and the system processor.
    Open Task Manager while you run your Speed Test, and see what kind
    of core-railing is going on.

    *******

    In this example, a very weird symptom. When he appears to be using some >Southbridge (PCH) lanes, the card seems to be resetting itself.

    https://www.reddit.com/r/pchelp/comments/1ko2xk3/tplink_tx401_constant_connecting_and_disconnecting/

    One commenter in a thread seems to think the chip can't drive a
    long cable, and there may be a length limit issue. My cable is CAT6A
    and is ten feet long, for whatever that is worth.

    On my machines, one machine is using the upper slot for the TX-401,
    the other machine is using the lower x4 slot for the TX-401. Makes
    no difference, here. No problems being cause by a slow choice.

    There is a Powershell

    https://stackoverflow.com/questions/66279388/any-windows-alternative-to-unix-utility-ethtool-for-autonegotiation-of-ethernet

    Get-NetAdapter

    and friends, a bit limited but something to try. But all the second
    command is doing, is displaying what's in that Device Manager panel.
    There still isn't as much info as I would like.

    While I could look in Eventvwr.msc, I don't know where exactly
    any errors from a NIC would end up.

    *******

    OK, there are some comments here that seem to match yours. Some
    people seem to be moderately happy. Linux always gets good press.
    Windows, not so much.

    https://www.pbtech.com/pacific/product/NICTPL1001/TP-Link-TX401-10Gbps-PCI-E-Ethernet-Adapter-1-x-RJ

    This is the Device Manager "Details" tab for mine, Hardware Ids.
    See if you have the same hardware as mine.

    PCI\VEN_1D6A&DEV_04C0&SUBSYS_00011D6A&REV_03 >PCI\VEN_1D6A&DEV_04C0&SUBSYS_00011D6A
    PCI\VEN_1D6A&DEV_04C0&CC_020000
    PCIIVEN_1D6A&DEV_04C0&CC_0200

    Paul



    Hello Paul. Yes as you.

    --- PyGate Linux v1.5.17
    * Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)
  • From Paul@3:633/10 to All on Wednesday, June 17, 2026 15:31:35
    On Wed, 6/17/2026 10:27 AM, Java Jive wrote:
    On 2026-06-17 13:06, Dan wrote:

    So, I can only fix this by a full re-install of Win 11?

    To amplify, I hope, the reply that Paul has already made ...

    He's not expecting you to throw away your current installation of Windows, but rather to back it up with an imaging program such as Macrium Reflect, the last free version of which is 8.0.7783, available here ...

    For completeness 32-bit version ...

    <https://download.macrium.com/reflect/v8/v8.0.7783/reflect_setup_free_x86.exe>

    ... but more probably what you want, 64-bit version ...

    <https://download.macrium.com/reflect/v8/v8.0.7783/reflect_setup_free_x64.exe>

    To do this you will need a spare hard disk of sufficient capacity to hold a back-up disk image, and a means of connecting it during the backup.

    So the procedure would be ...

    1)  Backup current install of W11 to some form of external storage;
    2)  Reinstall W11;
    3)  Run the same speed test again and note the results;
    4)  Depending on the results of Step 3, one of:
            Restore the original W11;
            Remake the original W11 by reinstalling in stages everything
            that was on it, testing the speed after each major installation,
            finally copying your data from the original installation.

    Of course, the above is tedious, but it may be the only way to get to the bottom of what's causing the problem.


    Since we're finding other people with TX-401 with the same problem
    (and with radically different achieved bandwidths), it's something
    on the Windows side that isn't working properly, but it's not clear
    what it is. Maybe this experiment, of trying a clean install of
    Windows, has already been tried by someone out there and it didn't help.
    I didn't see anyone report that though, but neither has anyone
    managed to fix it. I've not spotted any "[sorted]" reports.

    Paul

    --- PyGate Linux v1.5.17
    * Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)