Xiaomi AX3600 performance thread


I recently replaced my trusty Archer C7's with Xiaomi AX3600 as AP's and they perform very well on WiFi (having 4x cores and 802.11ax). I am almost able to saturate 1Gbit on LAN side via 2x2 80Mhz wide 5GHz 801.11ax device. iPerf3 is around 820Mbit both ways.

I haven't had time to test its routing/shaping/crypto performance though. But I would like to replace3 my current (aging) x86 router with another AX3600. So I hereby wonder if anyone was able to benchmark AX3600 as wired router (preferably using iperf3)?

  • NAT performance WAN <-> LAN (should not be an issue as CPU is plenty fast)
  • SQM performance WAN <-> LAN (using layer_cake)
  • OpenVPN performance (128, 256-bit encryption?)
  • WireGuard performance?
  • Any tuning tricks (using irq_balance)?


these two you should be able to test yourself, pretty easily, by configuring it as router again.
in the 2nd case, setting up SQM too (and temp disabling it on the x86).

My AX3600 are "in production" as AP's and it took a lot of tweaking to get VLAN's right. I should have done testing before but it is lot of fuss to do it now.

Please don't postulate results before having tested them, you may not be able to back them up with facts.

WG is tested here, single core though.

Except that the ipq8072a SOC in the DL-WRX36 is clocked at 4*2.2 GHz, while the ipq8071a SOC in the Xiaomi AX3600 is clocked at 4*1.38 GHz.


didn't look it up, apples and banas then :frowning:

If we assume linear performance scaling, 1.2GHz core should give us 160:ish Mbit/sec Wireguard performance. Does ipq8072a have any crypto accelleration instructions that Wireguard or OpenVPN can use btw?

Wireguard, no - but ARMv8 crypto support (usable via openssl) is present.

Is the performance better than the RT-AX53U though it has no hardware offloading?

That depends on what you're looking for. If hardware flow-offloading works, mt7621a can route at 1 GBit/s wirespeed - for anything that can't be offloaded, the 2*880 mips 1004Kc SOC will be struggling. ipq8071a has 4*1.38 GHz cortex a53/ ARMv8 cores, which is faster than mt7621a for pretty much any purpose (and the wifi is also considerably higher end, 4x4+4x4 vs 2x2+2x2), but it won't reach 1 GBit/s while routing ('thanks' to its very basic and unoptimized ethernet/ switch drivers).

1 Like

So the AX3600 doesn't have hardware flow offloading, even though there is a box to tick in LuCI?

It does not, the checkbox does not test for availability.

Can someone tell me the values of "driver default" dbm of 2.4G and 5G?

Are this values recognized by 5GHz driver? Is it possible to enable beamforming and BSS coloring? Or is it perhaps driver default for ath11k?

option he_bss_color '8'
option he_su_beamformee '1'

I have been running mine as AP for over a month on both 5GHz ax and 2.4GHz ax (guest). Between 5-10 clients.
As far as I can see, early issues with leaking memory are resolved. It is fairly stable (although memory hungry):. This is one month memory plot:

1 Like

Not in all cases. Mine is usually like yours, stable for several days/weeks, but from time to time it goes worse. It starts to eat memory, it goes over 85% or more.
I can reproduce it without problem. I have two IP cameras, if I disconnect them, the router starts to eat memory. If I connect them again, the memory returns to the earlier values and remains stable.
But sometimes, with the cameras connected, I have the memory problem and only a reboot fixes the issue.
For example, this night one of the cameras disconnect, and produced this:

Speaking of performance, does someone have a good DAWN or usteer config to share when using 2 x ax3600 connected via wired. it's odd that the device wouldn't connect to the nearby router even if i stand next to it. has anyone tried irq_balance? how is sqm helpful?

I did the following tweaks for better wired performance:

Network > Firewall > General Settings > Routing/NAT Offloading > Enable Software Flow Offloading (Default is Disable Hardware Flow Offloading>

Network > Interfaces > Global Network Options > Enable Packet Steering

1 Like

In my experience with both, proper configuration depends on multiple factors and configs aren't often "plug-and-play" between individuals. For example, proper configuration would take into account the signal level overlaps between your APs in your exact environment. Then various settings in either config would be adjusted based on those levels and your goals for band steering and/or AP steering.

What do your logs indicate were going on when your device wasn't connecting?

Are these configured as AP's? In that case, SQM is not relevant, neither is flow offloading (which will also disable SQM).

I use 2xAX3600 as AP's and my devices roam fine between them, despite 802.11r not being on. There is plenty of CPU left even when fully loading the AP, so I haven't bothered installing irq_balance.

1 Like