Backport r8127 driver for Realtek RTL8127 10G NICs

Problem/Justification

The Realtek RTL8127 is quickly becoming the standard low-cost 10GbE controller on new consumer motherboards (Z890/X870) and budget PCIe cards. Because TrueNAS 25.10 is on Kernel 6.12, and native support for this chip didn’t land upstream until Kernel 6.15+, these cards are dead on arrival.

Currently, users are stuck between two bad options: using older, hotter, and driver-buggy Aquantia AQC113 cards, or trying to manually compile DKMS drivers on TrueNAS, which breaks “appliance” stability and gets wiped on updates. This is the exact same situation we saw with the Intel i225/i226 launch; the hardware is here, but the OS is lagging behind.

Impact

Adding the r8127 module would enable out-of-the-box support for a massive wave of modern, power-efficient hardware. It saves users from having to return new motherboards or buy used enterprise gear just to get 10G working.

Disadvantages: None. It’s just a driver module.

User Story

A user installs TrueNAS on a modern motherboard or installs a cheap PCIe 4.0 x1 10G card.

  1. Currently: The interface doesn’t show up. lspci shows the device exists, but no driver loads. The user has to resort to the command line to hack in a driver or swap hardware.

  2. With this fix: The driver loads automatically on boot, the interface shows up in the GUI, and 10G just works.

Technical Note:

  • PCI ID: 10ec:8127

  • Driver Source: Realtek has an out-of-tree driver (r8127) that is DKMS-ready and can be integrated before the full Kernel 6.16 rebase.

Note: This is a replacement of the original feature request that I posted here with an error where I wrote 8727 instead of 8127,

1 Like

I can see the appeal to the CE crowd. On the enterprise side, there will be little to no interest for this driver. A used x710 NIC costs less than $50, can run at sub-5W.

Combine that with the LONG history of Realtek NICs being buggy, your proposal has a big hill to climb. I doubt anyone at ixsystems wants to “own” the associated issues if such driver / hardware issues affect this particular Realtek chipset.

Good luck!

2 Likes

I agree with you, most CE users don’t have an abundance of PCIe lanes and cards like the X710 take x8 of them. This takes x1, and even while not being as reliable as Intel, delivers reasonable performance for home use.

3 Likes

I doubt TrueNAS (the company) will add this driver and would likely wait until they are using the 6.15 kernel. I’m just saying to not get your hopes up with this request. This is how they do things and stability is critical, so dropping in a driver they must build, that adds the possibility of instability. Is it worth the risk to the paying customers?

But I’m not saying they will not do this, and I understand why you are making this request, I’m only saying I doubt it will happen until they are using a kernel that directly supports it.

1 Like

Thanks! Appreciate your understanding. I will keep my fingers crossed.

As you said previously, you could just compile it yourself however any upgrade would require you to recompile it until it was supported. It could be a viable option for the time if really needed.

I’m using a very annoying AQC113 (Marvel) card till then. If you think Realtek drivers are bad..

2 Likes

Just compile it by yourself. Write an automation. Compile automatically every time you update. Power-on setting commands. Load the driver.

1 Like

Kernel 6.18 is this year’s LTS. Asking instead that it be the kernel for 26.04 may be the way to go.

Install 25.10.0.1 and compile the driver you need yourself

Next update is 26.04 with a new kernel, you have the driver natively.

2 Likes

This argument is a bit odd because TrueNAS currently with Linux 6.12 supports Realtek nics unless ixSystems went out of their way to compile the kernel without the realtek drivers (which I don’t think is the case).

All that is being asked is a backport for the currently officially supported driver in Linux 6.15, it wouldn’t any more or less supported than any other network driver. No one is asking ixSystems to write their own driver, and no one is really asking ixSystems to “own it”. All that is being asked is to backport it and others are already doing this (i.e. package: add kmod-r8127 ethernet driver (851ea69d) · Commits · gtu / openwrt · GitLab with openwrt)

That being said, waiting for a 26.04 that is using Kernel 6.18 is what the likeliest option will be.

Best as I can tell, ixsystems tends to be pretty slow adopting new drivers in general, perhaps allowing them to be refined with time? It’s not just NICs, it’s everything.

See NUT, which is stuck way back at the dawn of time. The good news is that at least with NUT you can spin up a VM and host a more recent version in there, then poll it from the NAS.

That doesn’t seem like a viable solution for a NIC, so I guess you’re stuck either going into developer mode and porting in the driver yourself or waiting for a later version of TrueNAS to include the driver.

Apologies if I come off as prejudicial re Realtek but the old forum is littered with posts where the issue could be traced back to a Realtek NIC or driver. It’s entirely possible that they’ve become better choices for TrueNAS in the meantime.

There are more and more affordable consumer NAS appliances coming onto the market which are not strictly locked in to their vendor’s OS and can run TrueNAS. They don’t typically have expansion options for PCIe. I’d like to advocate on keeping TrueNAS CE as an option for them :slight_smile:

If it ever gets to the point that drivers for consumer NICs need to be removed from the TN kernel, maybe there could at least be a GUI option to load them in as modules, or to switch to an HWE-type kernel (with a little warning / risk acceptance dialog).

Hey - my crappy AQC runs just fine - outside of needing 15 commands every boot & a few GUI toggles to work at all!

…I think I see your point.

1 Like

None taken, just pointing out that the logic of “Realtek doesn’t have the best drivers” is moot/besides the point, unless ixSystems go out of their way to prune and removing existing in tree Linux drivers (which they don’t afaik).

In any case, the RTL8127 will become the defacto 10gbit NIC for any personal/homelab case in the coming decade. NIC’s that use this realtek interface will cost ~$30 (its now double but it just came out) and it uses 1-3 W of power and the raw cost for motherboards is a few dollars which means that you will find it in every motherboard with 10G as well as every pCIE card that you buy on Amazon and co in the future.

Or to put it differently, its going to be come the Intel for motherboard NIC’s (especially given whats happening to Intel now) and while its true that historically Realtek has been pretty bad when it comes to driver department, from what I heard this specific driver works absolutely fine (also in Windows). The biggest issue people are getting is they are accidentally ordering the pCIE 3.x version off Alibaba/Aliexpress which doesn’t have the full bandwidth (see Realtek 10 GbE USB Adapters might be on the way? | ServeTheHome Forums )

I have zero doubt that consumer motherboards will continue to feature Realtek NICs, as they already do. The default speed will just gradually increase from 1GbE to 2.5 and then 10GbE. Presently, the NIC speed depends on the price point of the motherboard, the nicer stuff can already run really fast.

On the workstation and server side of the motherboard business I expect intel NICs to hang on for a while longer. This is the segment that spends $$$ on IPMI and other features that consumers usually have zero use for. Those customers want zero surprises and for whatever reason, Realtek had plenty per the old forum.

It is also a segment that is more conservative re: jumping hardware architectures. See how long it took for Ryzen to make a dent even though it arguably offered a better price / performance ratio than Intel CPUs for years. Similarly, it’s only in the last year or two that true ARM workstation motherboards were released.

I do not doubt that large cloud providers like Apple had already been running ARM based motherboards for years (they design them, so they have 100% freedom) but for the general populace, like HA-SMR and HM-SMR, the ARM options are slim.

So Realtek will continue to dominate its “niche” of the motherboard market and Intel will do its best to defend its even smaller but likely more profitable niche in the prosumer and server end of the market.

1 Like

Nope on both. “Workstation” has long falled for Aquantia (even Supermicro, tu quoque…), and there are more and more Broadcom NICs on the server side (anything from AsRock Rack which ends in “/BCM”, ranging from Ryzen to Ampere through Xeon, not to mention this Threadripper workstation motherboard from :scream:).
The AI rack belongs to Nvidia, by virtue of Connect-X NICs coming bundled with the GPUs. Intel doesn’t have anything to offer at 400+ Gb/s anyway

Intel “doing its best”—which is not that much these days—will NOT be enough.

2 Likes

I happily stand corrected.

Not sure where Broadcom stands re: NICs, are they stable / functional or not?

As for Aquantia, just today I read another forum thread re: stable performance in general but unstable, unexplainable behavior during boot or whatever.

On that aquantia-equipped motherboard you referenced, I doubt the aquantia chip is expected to host a 45-drive storage cluster (the CPU is gargantuan!). Rather, that board is designed around massive CPU loads and oodles of PCIe expansion.

If It was designed to be a file server, it likely would have a dedicated NIC port or two for VLANs, etc. The servers I have been looking at have either Broadcom (MSI D3052) or intel (MSI D3051) NIC chipsets, if they have modern NIC chipsets at all. One of the trends now for pro Ryzen motherboards seems to include “bare” motherboards with just a bunch of exposed PCIe lanes (h13sae-mf, for example) and the barest of infrastructure.

That intel has improvement potential is beyond a doubt. They got fat, dumb, and happy from dominating the CPU business as long as they did. Another great illustration of why monopolies are bad for business, including that of the monopolist (in the long run).

Not to mention that workstation boards may be expected to run Windows, or at the very least a desktop version of Linux, rather than a server OS…

How Intel intends to improve: :man_facepalming:

1 Like

I am very happy that we are discussing this and not fixating on expecting home users on CE to have Intel NICs, but the request still stays at 3 votes. Just saying.

That being said, yes, the AQC113 NIC is “fine” for the most part, and also comes in that PCIe 4.0 x1 option but I have genuinely battle tested the r8127 driver on other distros and it is much more stable and consistent than the AQC113 driver.

Just my personal 2 cents,

Who wins the race to the bottom segment of the 10G market is not really relevant, nor is even my personal distaste for Realtek NICs. TrueNAS updates its kernel yearly-ish in line with the launch of a new LTS kernel.

Users building NAS’s should review the hardware makeup of their choices and compare that to the current LTS kernel when making purchasing decisions. If driver support is not included for your hardware in the currently shipping kernel, it’s not really a TrueNAS issue. Since this particular card seems to have it’s drivers in 6.16 https://www.phoronix.com/news/Linux-6.16-Realtek-RTL8127A

Support in TrueNAS should come when 6.18 ends up in the product. This situation is in many ways similar to Intel ARC Battlemage GPUS, and the resolution for those users has been the same.

3 Likes