Has anybody compared real-life ethernet performance (with NAT and many firewall rules) of hAP ax2 and hAP ax3 in comparison to hAP ac2?

Is there a real-life performance benefit of ax2 over ac2?
And is really ax3 only around 30% faster than ax2 (despite up to 1800 MHz CPU vs only 864 MHz CPU)?

P.S. I have seen the tests (here, here and here).

I have no reason to doubt that the AX3 is best followed by hapac2 followed by hapax2 based on those tests.
Your in Riga, pop over to MT to confirm!!

Performance differences have never been a 1:1 function of clock speed for a system.

The ac2 actually gains in some cases over the ax2 by virtue of handling half the data width for an instruction, but I see no point other than marginal cost in going for 32 bit not 64, 128/16MB rather than 1GB/128MB.

The ax3 also delivers a 2.5GB port and external antennae for better radio sensitivity. The next step would be an RB5009 with wAP ax or 3rd party wireless.

Be aware that the hAP ac² numbers are RouterOS 6 numbers and you'll need to deduct 25% or so when doing comparisons with the two other RouterOS 7 (out of factory) devices. Look at the hEX RB750Gr3 as an example, before the hEX refresh was announced the old hEX boasted values of 385 Mbps in that special column https://web.archive.org/web/20241004221 ... estresults because those were RouterOS 6 test numbers. And now MikroTik has significantly reduced those numbers, by using the real applicable numbers from RouterOS 7 (to compare with the hEX refresh).

With my hAP ac² if you disable fasttrack, by using the defconf firewall with IPv6 prior to 7.18, or by using the defconf firewall for IPv4 but with the fasttrack rule disabled, I could achieve speedtest.net result around the 7xx Mbps range with RouterOS 7 (with PPPoE WAN). Which means it's inline with a 25% reduction compared to the published value for RouterOS 6. Which means the hAP ac² when running RouterOS 7 without fasttrack should be slower than the hAP ax² with its 913Mbps published ROS7 number. With the RAM and storage restriction, the hAP ac² should no longer be purchased new. If you can find it cheap on the second-hand market then you can use it as a wired-router/managed switch (turning on WiFi is risky because you can expect any new release of RouterOS 7 to completely fill the flash storage any time).

Never seen an explanation but read often, so I ask now: how come that v7 is slower than v6?

It's slower in those particular kind of benchmarks because in RouterOS 7 the route cache is no longer available. You can read more from this thread, especially the posts by @raimondsp:

viewtopic.php?p=882867#p882429
viewtopic.php?p=882867#p882867

So it really means that if you want to compare hAP ac² and hAP ax² numbers from the links in the OP, you'll need to scale down the RouterOS 6 numbers of the hAP ac² accordingly if you want to use that router with RouterOS 7.

Thank you for the answers!

I am seeking mostly experience stories (ideally somebody who upgraded from ac2 to ax3 or something like that)...

Regarding RouterOS 6 vs 7 speed on hAP ac2 - with RouterOS 6 this device has default CPU frequency of 716MHz, however the same device with RouterOS 7 has "auto" CPU frequency which can increase up to 896MHz under load. So hAP ac2 actually seems to have about the same speed on both RouterOS versions (maybe even a bit better on RouterOS 7).

P.S. I already have a few hAP ac2 devices and I am considering my options regarding the diskspace on ac2 devices with RouterOS 7 - currently I have around 300 KB free space (with routeros and wireless packages) and I am hoping to be able to update to 7.19 or maybe 7.20, but I need to find a long term solution. I am considering to either mod my ac2 with RouterOS 7 devices with bigger flash memory (as described here and here) or less likely buy some other devices like hAP ax3 (I am using switch rules, so in case of upgrading to ax3 I would need to make another bridge and take a performance hit - this makes me a bit skeptical about such upgrade).

For the hAP ax² and ax³ currently there is no performance difference if you create additional bridges, because there is no hardware offload to lose. Usually with the other devices only one bridge can be hardware offloaded, and MikroTik advice is normally to only create one bridge per switch chip. But on the hAP ax² and ax³ currently everything runs on the CPU anyway because hardware offload is not yet supported. The bridging numbers published by MikroTik are all non-hardware-offloaded numbers.

My comparison with fw + nat + mangle + ipsec shows that ax2 performs worse than ac2 or ac3.
You shouldn't consider of buying ac2 due to insufficient flash size for ros 7 (16mb)
Only ac3 (128mb).

My humble opinion: ax2 cpu arm64 surely faster than ac2\ac3 cpu arm32.
But ros 7 not optimized for ax2 better performance.
So right now ac3 performs faster.
And 4011 faster than ax2 \ ac2 \ac3 for my tasks.
Consider of buying used 4011.

viewtopic.php?t=214944