Realtek 10 GbE USB Adapters might be on the way?
- Thread starter jerrythea
- Start date
Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.
When the likely issue is 3.2 gen2x1
It was explained in a previous post already, they are implying that the lower performance is a windows overhead thing, not a USB branding/confusing thing of USB4 not always supporting gen3.2x2 performance due to the change in architecture to thunderbolt.
The marketing material is happy to let customers have poor performance.
If you demand information, while ignoring what has already been said, you might frustrate your sources of information.
I know you may feel strongly about those that are poor receptors of information... but tone down the harshness...
Not true
I have tested my RTL8159 on three different USB4/Thunderbolt ports, all 3 gave me the full 20Gbps aka 2 10Gbps lanes
2 ports based on Intel and one from AMD
So the marketing is actually not that misleading at all..
Does USB4 ports exists that doesn't support gen3x2 proberly yes, but it far from all that dosen't
No, it is not a mandated part of USB4 spec.
Yes the marketing is implying bad performance is a driver issue and not a port issue.
7Gbps or worse is due to 1 lane tunneling, not a driver, that is bad information and outright misleading.
I am not sure why you are choosing this hill to die on, did you write this marketing material?
Devices with USB4 that do have the higher speed ports often have other ports that are only 10 or 20Gbps... connecting to those would cause issues that might be unexpected if the marketing only says gen3.2x2 for best performance and a port labeled USB4 20Gbps can still be 1 lane... and not give the required performance.
Last edited:
In my anecdotal experience with thunderbolt 3 and 4 controllers, most of them didn't do USB 3.2 Gen 2x2. Also USB4 is based on Thunderbolt3 but is not Thunderbolt3.
I think the marketing copy of Cable Matters is not horrible and this mess is not their fault. The say typical 7 Gbps windows because I am sure they terrified by the returns.
I think the marketing copy of Cable Matters is not horrible and this mess is not their fault. The say typical 7 Gbps windows because I am sure they terrified by the returns.
I can only report what i see, I will happily do some testing this weekend
Here is a thought, since USB4 gen2x2 is also 20Gbps maybe the firmware in my Diewu also implements this link mode besides the "normal" USB3 gen2x2
I'm not discussing iperf/driver performance, just the actual USB link layer
Last edited:
So long as it has 2 lanes it is fine.
My Zeph G16 has many USB4 ports, but only 1 is 40Gbps and guaranteed to have 2 lanes. The 20Gbps ports could be 1 or 2 lanes.
And the 10 is most certainly 1.
This is just a repeated headache brought on by USB-IF being absolutely terrible at their naming conventions as well as an architectural shift to thunderbolt. And requires extreme clarity in marketing material to avoid blowback from unexpected behavior and they are lining themselves up for it by advertising thunderbolt support when it is not a native thunderbolt device and requires 2 lanes to work at full speed.
They are trying to mitigate it by labeling 7Gbps as expected performance. We got 9Gbps in our lab but you can only expect 7... because windows and drivers. It is misleading even if the origin of the chaos isn't their fault.
Last edited:
That is technically USB4 V2 on that box... so its good that it works at proper speeds lol.
No only the USB 4 ports are able to do USB 3.2 Gen2 x2
The other ports are USB 3.2 Gen2 ie 10Gbps before usb overheads so 6.7Gbps of throughput
I have looked into this further as i havnt really done a deepdive since Thunderbolt 5 came out
USB 3.2 Gen2x2
Intel Thunderbolt 3 Controllers = No
Intel Thunderbolt 4 Controllers = No (edit Metorlake embeded Thunderbolt 4 adds support back in)
Intel Thunderbolt 5 Controllers = Yes
ASMedia ASM4242 = Yes
Realtek RTS5490 = Yes
Since ASMedia make the PCH for AMD depending on the generation of IP ASMedia Used it may or may not do 3.2 Gen2x2 ie the newer it is the more likley it will be based on ASM4242 IP
Your MinisForum Device's USB 4.0 Ports are implemented with an Intel Thunderbolt 5 Controller but havnt been certified by MinisForum so they cannot call it Thunderbolt 5 so yes USB 3.2 Gen2x2 works fine on it
Last edited:
Hi,
anyone have working RTL8127AF with SFP+ DAC passive or active AOC have link on linux?? I try everything with no succes..still no link :c
With Windows work good with setup in driver settings 10G Full Duplex
My setup:
RTL8127AF XikeStor SKN-8127AF-1SFP+ aliexpress.com/item/1005011607908697.html
Proxmox 9.1.6 6.17.13-1-pve
Cable:
10G SFP+ DAC (3 diffrent)
10G SFP+ AOC
Drivers: 11.016.00 from Realtek website
Switches for TP-Link Omada:
SG2210XMP-M2 v1.0
SX3008F v1.20
anyone have working RTL8127AF with SFP+ DAC passive or active AOC have link on linux?? I try everything with no succes..still no link :c
With Windows work good with setup in driver settings 10G Full Duplex
My setup:
RTL8127AF XikeStor SKN-8127AF-1SFP+ aliexpress.com/item/1005011607908697.html
Proxmox 9.1.6 6.17.13-1-pve
Cable:
10G SFP+ DAC (3 diffrent)
10G SFP+ AOC
Drivers: 11.016.00 from Realtek website
Switches for TP-Link Omada:
SG2210XMP-M2 v1.0
SX3008F v1.20
I also have no issues under windows with RTL8127AF, but when placed in MS-02 Ultra, It broke both onboard RTL8127 (RJ45 version) and RTL8127AF.
With both installed i was not able to get link on either..
A night of troubleshooting came up empty.. It seems support for SFP+ version is very limited under linux atm even with HWE Kernel/latest driver from Realtek
After testing all the various versions of RTL8127 and beyond ..
I ended up using my trusty X710-DA4 ... and all my troubles just vanished
Last edited:
@Shonk
On the MS-02 I have both a USB4 v2 controller
82:00.0 PCI bridge: Intel Corporation Thunderbolt 80/120G Bridge [Barlow Ridge Host 80G 2023] (rev 84)
And a USB4 v1
00:0d.0 USB controller: Intel Corporation Meteor Lake-P Thunderbolt 4 USB Controller (rev 10)
Both support 20Gbps
So
Intel Thunderbolt 4 Controllers = No
is not true for all im my testing
But in fairness a very new implementation of T4
On the MS-02 I have both a USB4 v2 controller
82:00.0 PCI bridge: Intel Corporation Thunderbolt 80/120G Bridge [Barlow Ridge Host 80G 2023] (rev 84)
And a USB4 v1
00:0d.0 USB controller: Intel Corporation Meteor Lake-P Thunderbolt 4 USB Controller (rev 10)
Both support 20Gbps
So
Intel Thunderbolt 4 Controllers = No
is not true for all im my testing
But in fairness a very new implementation of T4
Last edited:
There’s an open report on the Linux netdev mailing list from November 2025 that describes what looks like the same issue: RTL8127AF detected fine, no link over SFP+ DAC.
netdev: RTL8127AF doesn’t get a link over SFP+ DAC
In that case, the problem was the mainline r8169 driver, which doesn’t properly handle SFP+ link negotiation on the 8127AF. It reports the port as “Twisted Pair” even on the SFP+ variant and can’t disable auto-negotiation. The reporter resolved it by switching to the out-of-tree r8127 driver from Realtek, which brought link up immediately at 10G full duplex.
Now, you mention you’re already running the r8127 11.016.00 driver from Realtek’s site, so the obvious fix doesn’t apply here. A few things worth checking though:
First, verify the out-of-tree driver is actually the one bound to the device. Run ethtool -i <interface> and confirm it shows driver: r8127 and version: 11.016.00. On Proxmox, the mainline r8169 can race the out-of-tree module at boot and grab the device first, especially if r8169 wasn’t explicitly blacklisted. If you see driver: r8169 in that output, the Realtek driver compiled fine but never got to claim the card. Blacklisting r8169 in /etc/modprobe.d/ and rebuilding initramfs (update-initramfs -u) would be the fix there.
If it is confirmed r8127 and still no link, check what dmesg | grep r8127 shows. The netdev reporter’s Realtek driver output showed “Supports auto-negotiation: No” on the SFP+ interface, meaning the DAC connections are forced rather than negotiated. In their case link came up automatically. If your dmesg shows the driver loading but no “link up” message, try forcing it with ethtool -s <interface> speed 10000 duplex full autoneg off on the r8127 driver and see if that changes anything.
Also worth posting the output of ethtool <interface> with the r8127 driver loaded. Specifically what it shows for supported link modes, port type, and auto-negotiation. If it’s still reporting “Twisted Pair” and “Auto-negotiation: on” with the Realtek driver, something may not have compiled or loaded correctly.
Sounds like it could be a platform-level problem rather than a driver issue. PCIe resource allocation, IRQ conflicts, or BIOS limitations are the usual suspects when two devices that work independently stop working when both are present.I also have no issues under windows with RTL8127AF, but when placed in MS-02 Ultra, It broke both onboard RTL8127 (RJ45 version) and RTL8127AF.
With both installed i was not able to get link on either..
A night of troubleshooting came up empty.. It seems support for SFP+ version is very limited under linux atm even with HWE Kernel/latest driver from Realtek
After testing all the various versions of RTL8127 and beyond ..
I ended up using my trusty X710-DA4 ... and all my troubles just vanished
Worth checking dmesg | grep -i r8 and lspci -vvv for both devices when they’re both installed to see if the kernel is even enumerating them properly. If one is getting starved for PCIe resources or there’s an IRQ conflict, the errors should show up there. Also check if the MS-02 Ultra BIOS has any settings related to PCIe lane allocation or bifurcation that might be relevant. Were you running the mainline r8169 or the out-of-tree r8127 driver when both cards were installed?