Hi 500/500 is a very max for this router
fq_codel instead of CAKE
2 Likes
@jmandawg With CAKE SQM on my Belkin RT3200, I got max 450 Mbps (or may be even lower, around 420 Mbps, don't remember the exact number).
For higher throughput, you need to use fq_codel. Not sure what's the max limit is with fq_codel on RT3200 though.
6 Likes
SQM uses a lot of CPU but so does wifi on this router, so wouldn't that create a choke point if some wifi device is using a lot of bandwidth? I guess that makes the case for using a dumb AP and standalone router.
2 Likes
has anybody investigated on if we can overclock the cpu core a bit to squeese a little more speed to hit that magical 500Mbit with cake
Overclocking does not make sense for this MT7622 device. It will still be dual core ARMv8 Cortex-A53, designed for high energy efficiency, not for highest performance.
Overclocking is guaranteed to shorten the lifetime of the hardware and lower stability even while the peak power is not needed.
If you can’t keep your hands away from overclocking: at least build a more powerful active cooling solution with a fan, to lower temperatures.
If you need a more powerful router, I suggest to instead upgrade to Filogic 830 MT7986 devices with 2.0 GHz quad core Cortex-A53 instead. Or even MT7988 devices with Cortex-A73 cores.
given the routers age I would say run it to the moon if it pops it pops ...
the question is HOW ? does the uboot need to be patched or is there a easier way
For some reason your post reminded me about this old video:
Olden but golden!
In respect of the two categories shown in the video, namely: the Intel CPUs gracefully handling heatsink removal; and the AMD CPUs going up in smoke upon heatsink removal, I suppose this router CPU would fall into the latter category? Perhaps someone could confirm?
Were it not a shame to risk frying something for the sake of entertainment, I’d suggest that you make a video for us @senposage!
IT takes a LOT to damage any modern processor with overclocking you need to actively try and kill it.
the MediaTek MT7623 in the Banana pi R2 is widely known to overclock decently
if the MT7622bv follows the same trend then one might be-able to modify the
dtsi/dts and rebuild
haven't found the clk definitions yet still looking ...
Is that based on experiences with other hardware than the RT3200? Sounds like quite a big overclock compared to the 1.3GHZ(?) default.
it seems like the mt7622.dtsi is missing from the tree
there is a patch referencing it /arch/arm64/boot/dts/mediatek/mt7622.dtsi but I do not see that path on the github unless its pulled from the main kernel sources at build
the max voltage appers to be
1350000
(ref)
it should just be a matter of changing the opp to whatever
then setting the maxium via the govenor at boot
sadly I am working from a laptop while my main system is down so I am not setup to build this and try it
at 1.35v I would expect that 1.5Ghz would not be out of the question for 90% of samples (likely much more but lets be conservative )
opp-1350000000 {
opp-hz = /bits/ 64 <1350000000>; /
opp-microvolt = <1310000>;
};
};
edit: good lord whatever software is powering this board is horride please upgrade to something written in this decade YIKES!
Wrong. The power consumption and temperature stress increases exponentially with higher clock speeds: to be stable with higher clock speeds you often need higher voltages. For your statement to be correct, you have to at least significantly improve the cooling: active fan cooling or even liquid cooling.
The whole effort of overclocking MT7622 with dual core Cortex-A53 1.3 GHz is useless since there are affordable MT7986 quad core Cortex-A53 2.0 GHz and MT7988 quad core Cortex-A73 on the market with OpenWrt support. It does not make sense to torture MT7622 devices without better cooling and publish ways for possible small performance increases and lower stability when you could instead just upgrade to newer more powerful devices with perfect hardware stability on stock specs.
Overclocking could make sense when you are already on the fastest available stock hardware on the market, you are improving the cooling and your application does see a significant benefit from overclocking on this already fastest available hardware.
Your idea of
should not get support here because harmful.
5 Likes
I agree, it's not worth it to overclock this SoC and get only a bit of performance at the cost of heat and decrease the life of the device.
It makes me wonder (and in the most respectful way): what do you guys do to saturate the Gigabit bandwith???
I have 1Gbps fiber and sometimes download things, have IPTV, make videocalls, play Fortnite (albeit not in a competitive way) and everything is smooth. I even share WiFi with a neighbor.
I do have my RT3200 with Packet steering on and the Flow Offloading enabled, HW and SW (of course, no SQM).
However, I don't know how to get gigabit speeds over WiFi. I have a laptop with an AX210 and I get up to 700Mbps download and up to 800Mbps upload. With ethernet, I do get close to 1000Mbps.
Another question? How do I enable irqbalance? I read that is most useful for quadcore processors. I also read that you need to tweak it to separate loads. But How do I do that???
Again, you know a lot more than I do and I mean no disrespect.
Cheers!
it really sounds like you have no experience overclocking things fair but lets break it down.
You are talking device life Mesured in YEARS Possibly being reduced by a small Fraction
read the datasheet here: https://download.kamami.pl/p579344-MT7622A_Datasheet_for_BananaPi_Only(1).pdf
(yes its for the A varient but for intents its the same chip) the b/bv lacks the video engine but otherwise the same core.
package TJMAX is a wopping. 125c ... you get anyware near that you are probly in deathvally at high noon running sqm benchmarks for your starlink ...
tho I would recommend keeping it under 80c for overclocking... temp x current draw is what induces damage not just temp. or frequency or powerdraw
the worst that could happen is a brick due to instablity. which is why we always boot at stock frequency then enable the oc by changing the govenor prams at boot with a script
but would wanger anybody moviated or talented enough to get Openwrt running is going to be aware that here be dragons and without some modifications at a glance this is going to require a recompile anyway.
opkg update;
opkg install irqbalance;
uci set irqbalance.irqbalance.enabled="1"; uci commit irqbalance; /etc/init.d/irqbalance restart
hardware interrupt distribution can be viewed via
cat /proc/interrupts
Thank you!
I was aware how to install it, but I'm not sure on how to configure it.
I read somewhere that I need to see the interrupts and set ones to CPU0 and others to CPU1. It's not automatic, right?
I posted the necessary configuration steps above:
1 Like
there is a luci package for irq balence
You can install luci-app-irqbalance
edit: oops older post, somehow this was at the top of my feed.
What does this offer over irqbalance I wonder?