RTL8125 2.5GbE Ethernet Controller (rev 05) - 2.5Gbps link speed not working (r8169 module)

[rbirkner (Rafael Birkner)]

After Upgrading from 1GB to a 2.5GB switch, the Ethernet connection is so slow that it appears offline to applications on Kernel 6.18 which is supposed to be stable .

At first all looks OK:

What I’ve noticed is the fact that too many packets are being dropped after the interface is up:

then even ICMP drops a lot;

Attached to 1GB port everything works normally with full speed.

Websites are unusable as well:

To check if there is no HW-issue I booted up from a live .iso, there the adapter is working normally with 2.5GB full speed with older kernels though (fast.com shows 2600mbps down and 125mbps up)

I must assume something in the driver or firmware broke over time…,

Any help would be appreciated!

[computersavvy (Jeff V)]

Note that you appear to be running rawhide (currently f44) which is a development version and issues there probably cannot be helped on this forum which is for the currently supported and released versions (at present f42 & f43).

The kernel you show (6.18.0) is actually behind the version that was just released for f42 & f43, which is 6.18.3.

It is quite common for users to encounter problems when running rawhide since that is constantly in development flux and stability is never guaranteed.

[Espionage724]

I wonder if a Debug kernel could affect speeds? I’m thinking the Live ISO would also run a Debug kernel though, but maybe there’s a kernel version difference from installed vs Live? (rawhide has a no-debug kernel iirc; a debug kernel on FreeBSD had ~20MB/s difference on LAN vs no-debug)

[rbirkner (Rafael Birkner)]

This issue has nothing to do with rawhide, all kernels are affected, just booted up

slicer@cosmos ~> uname -a
Linux cosmos 6.18.3-200.fc43.x86_64 #1 SMP PREEMPT_DYNAMIC Fri Jan 2 20:10:56 UTC 2026 x86_64 GNU/Linux

no change, the label is misplaced

btw: also the most recent CachyOS has the same issue, so this is probably not even related to fedora, but who I should ask Lunus himself?

[rbirkner (Rafael Birkner)]

it’s not about slow transfer rates, this is stall of data flow to a point that all traffic is halted…. I can also show that old kernels are working perfectly in this regard if someone cares.

[rbirkner (Rafael Birkner)]

I did also a test on Ubuntu, please compare:

[rbirkner (Rafael Birkner)]

And finally I wanted to show that older kernels work flawlessly (but I cannot downgrade to them as they don’t support my gfx-card yet as you see…)

[pg-tips (P G)]

@rbirkner

btw: also the most recent CachyOS has the same issue, so this is probably not even related to fedora, but who I should ask Lunus himself?

You can raise a ticket in Red Hat Bugzilla, and the Fedora kernel maintainers should report it upstream if necessary.