Sub 200€ router with mature support and good throughput on 1Gbps connection

@moeller0 @diizzy @slh @Borromini -- this thread turned into a great read thanks to you gentlemen. :wink:

RockPro64 does look intriguing, wish I knew about it when I placed an order at pine64 a little while back.

Do share more about your setup -- did you have to print your own case? Is there a multi-port NIC you can recommend?

I current don't have a good case for it however

I deployed several NanoPi R2S (RK3328) in December, each with Armbian and two OpenWRT virtual machines on top, and I don't have a single performance complain. With taking difference between RK3328 and RK3399 into account it's easy to guess that NanoPi R4S should be much faster (I use RK3399 on several other boards for one and half year - two additional Cortex-A72 makes a difference) although base config with metal case costs two times more.

You can find temperature and frequency tests here:

Hey guys, thank you all for your replies! I think this is really sad. To be honest, when I got 1Gbps cable a couple of years ago, that was the reason for me to leave OpenWrt. Because my old WR1043ND v1 and WDR4300 (which I only used as a wired router) could not handle throughput.

Right now I am pretty close to your recommendations, I am using a wired Ubiquiti ER-X for Routing,Gateway,DHCP etc. and an Ubiquiti nanoHD as AP. Both with original Firmware not with OpenWrt.
I thought it would be nice to have another look around and check if there was an open source alternative but it seems there is not.
To be honest SBCs don't seem to be the solution. Some like NanoPi etc have shitty community support, lacking documentation, old kernels etc. Others like RPi have slow i/o, high power draw and saved money at the wrong end (like crypto extensions). All have in common that their NICs are not great and adding at least one NIC by USB doesn't seem a good option anyways. And what good is an open source firmware, when hardware is largely undocumented.

So it seems I will go x86 sometime in the future. To bad Intel has been holding out on new Atom CPUs for almost 4 years now. But with the upcoming Jasper Lake Atom CPUs (N6005, N5105), I might give x86 another shot if price/performance/powerdraw are as good as I hope. (I know that Intel CPUs are far from open source hardware too, but at least support and documentation should suffice to get a decent router running without problems)

Mikrotik RB3011 with experimental OpenWrt. Even has an SFP slot for your fiber. I'm now trying the NSS accelerated stuff as well. Looks real promising...

1 Like

You can run OpenWrt on the ERX.

Not sure what you are talking about here - you can use both of NanoPi R2S and R4S with Linux 5.10 (Armbian images). What information you can not find in vendor or Armbian documentation?

On NanoPi R4S second Ethernet port is wired via PCIe, it was mentioned before here. I wonder if you actually read the specs or at least this topic?

What exactly you been missing here and here? What part of this hardware is undocumented?

1 Like

If you're fine with having separate APs, an Odroid H2+ might also work.

Somewhat related: I think it is time for a community effort to create some kind of overview for performance numbers of at least the most common targets/SoCs.

Edit: Never mind, forget I said anything. Obviously nothing short of a million dollar lab for testing and down-to-the-last-bit reproducible values would do. Let's keep answering "what router should I get" threads individually. Over and over. And over.

1 Like

While I'd really love to see these numbers, benchmarking is hard - even harder if it's supposed to be comparable between different targets and devices.

Simple complication, PPPoE (which some ISPs even insist on using for FTTH at high speeds), which has a quite major impact on performance. If you've selected your device well, you'll stay inside the performance abilities of your router - and can't really do real-world testing against your ISP connection. Sure, one could set up a PPPoE server locally, to run iperf over that local PPPoE connection, but doing that is non-trivial (and to be honest, that would be too much hassle for me).

Similar issues arise with SQM, I wonder how relevant local testing over symmetric ethernet could be, where SQM doesn't really have to do much - compared to the ISP connection from hell, with huge buffers, highly asymmetric throughput and normal ping times.

Yes, a lot of this can be done, but reproducible benchmarking is hard and I fear too hard for most of us (including myself). @jeff has tried to come up with this (in synthetic benchmarks) a while ago, but I wonder if anyone else can actually validate these figures on their hardware (no, I don't doubt his testing or figures at all, but slightly different test procedures might result in quite different ones). Anecdotal values (real-world experiences) are another topic, they are valuable, but they usually won't help those who need these estimates most, those rare few who really push towards the magic 1 GBit/s barrier.


I managed to confirm his results for SQM on an turris omnia, so I believe the results are trustworthy, but especially the VPN tests are somewhat hard to reproduce (in an production network one might not necessarily want to change all IP addresses "just" for a performance test). I love how skilled & careful @jeff was in setting this up and in describing it so that it did not leave wrong impressions about what was tested and how it generalizes.

1 Like

Great that you could confirm those figures. Yeah, doing these tests well (as well as he has) is hard, especially if you're working with a single specimen that's supposed to be in production and don't have a lab environment pre-configured and waiting for action.

1 Like

At one point I worked on some data collection scripts that would look at percentage CPU idle and could collect statistics through time. With appropriate adoption and running it long enough, we could organically estimate the realistic "fastest reliable speed" for given routers... But the interest level wasn't that great, and the scripts weren't that well developed.

Realistically what's needed is cash. For a couple hundred $K you could get someone to set up a lab and buy one of each of say 25-50 of the top devices, and run a comprehensive set of tests in a test bench... In the end, I doubt such resources are available.

Also, for those who really care about performance, we already have the answer:

  1. x86
  2. RPi4
  3. NanoPi r4s
  4. RockPro64
  5. A couple other devices....

Plus commercial APs


IMHO: Linksys WRT1900AC[S] (~$50) + Ruckus R610 (~$150) on eBay

Hi everyone, this subject is really interesting, because I will soon switch to 1000/1000 debit gigabit fiber

I have a mikrotik hap AC2 under openwrt still under construction,

simple question ?

how much throughput can the SQM go when enabled with this router and is it ok for gigabit fiber?

I formerly had rb750

but hap ac2 + wifi is newer and better at fiber, what do you say?

Capture d’écran 2021-02-19 à 20.34.18

100 Mb with cake

i think you are wrong because my friend has also 300 300 with sqm :frowning: but i have not tested with up connection )

106 Mb its the max with CAKE.