Power usage with latest Kernel 6.12 Server Ubuntu build

[Thewonderer]

Hi,

 

I've come from Joshua's 24.04 Server build.  I was concerned that the support was dropping off as he's having a well deserved break.

 

I've installed the latest Server build from here.   https://dl.armbian.com/orangepi5-plus/Noble_current_server-kisak

 

Works fine, have moved it to EMMC and removed the SD card.

 

I use a USB power meter to see how much juice it uses.

 

I noticed this build uses about 20 to 25% more when idle or under load compared to Josh's 6.1 kernel build.  I am suspecting 6.12 has enabled more hardware in the OPi5+ and this is the reason.

 

Is there a way to disable some of the HW to lower the idle power usage?

 

On average it uses 5w on idle.

 

Thanks.

 

Loving the Armbian build. 

[eselarm]

Some time ago I also that various powermeasurements on my NanoPi-R6C and lowest was 1.5Watt. Was with NVME removed and kernel 6.1.75-vendor, 1Gbit GMAC ethernet. With CLI-tool 'btop' it is easy to see that idle lowest clock is 408Mhz.

With kernel 6.10.10-manline, the lowest clock is 1000Mhz, AFAIR idle was then 2.3Watt. Also checked with NVME inserted again, that adds another 0.3 Watt (when NVME is idle, uses way more when stressed a lot). This NanoPi-R6C is RK3588S SoC, OP5P+ is RK3588 and has more peripherals I think.

Used OS was Debian Bookworm based.

 

Just a quick check now with 6.12.5-mainline: 2.24Watt idle, clock 1000MHz. This is with Samsung 970 EVO+ NVME connected and OS running from it.

[Igor]

2 hours ago, Thewonderer said:

the support was dropping off as he's having a well deserved break.

 

I think support was dropped because:

- asking from him/us 1000 x more then he/we could deliver,

- users complete absence of reality (tl; dr;),

- users egoistic perspective as primary objective, 

- usage off private communication channels

- constant fight to get attention they don't deserve to get, 24/7 (from on person that needs to have its own life),

- asking amateurs to do in super limited time what pros in months or years can't,

- constantly attacking with "small" questions that requires hours or weeks of research in order to answer, months to fix some problem (users expect 1day resolution)

- most of damaging-time-wasting clients are powered by Dunning-Kruger, so the loss of our super precious time is ridiculously high,

- complete absence of financial support from amateurs,

- complete absence of financial support from professional abusers of open source (some HW vendors and their clients are not),

- constant pressure, stress and no emotional rewards whatsoever,

Even users treated him like shit and act abusive, same as here sometimes, he tried to be polite and professional. I don't known anyone that could endure constant insults very long while paying (!) for the time taking them. Imagine being a party host - how long you will stay professional when people f* with your family and puke around your condo? And how long you will host such parties? This is how this relationship looks like, if we allow your frustration to be in charge.

 

I understand that bugs in software are terrible thing, I also feel bad when this happens to me ... but can't pin bugs to us or requests features if you have no idea what you are asking for. We are minimizing this problem for everyone, to find sane grounds.

 

2 hours ago, Thewonderer said:

I've come from Joshua's 24.04 Server build. 

 

Welcome!

Joshua Ubuntu and Armbian Ubuntu are HW wise identical. He was contributing to the kernel our team maintains and we also supported him financially within our possibilities. Those two distros are different in user space / cosmetic details. We don't try to be Ubuntu, he tried to be Ubuntu. We removed Ubuntu stuff, snap is not preinstalled, there is ... I see value in Ubuntu packages as they tend to be more recent and polished then Debian Stable. Similar user-space philosophy as Linux Mint in PC world. This might give some insights https://docs.armbian.com/#comparison

 

2 hours ago, Thewonderer said:

compared to Josh's 6.1 kernel build.

Compare his 6.1 to Armbian 6.1 and you won't see any difference. Mainline, 6.12.y kernel is totally different thing, it has been developing more or less from scratch for more then two years. And is not fully completed AFAIK.

 

2 hours ago, Thewonderer said:

Is there a way to disable some of the HW to lower the idle power usage?

Very simple   Use kernel 6.1.y

 

If you find the way with 6.12.y, kindly share this information with others. Perhaps someone knows and will tell this. I don't know.

[Thewonderer]

Thank you Igor and Eselarm.

 

I saw some of the github issues Joshua had to deal with.  Glad he's looking after himself by walking away.

 

I believe Kernel 6.1 is the Android Kernel Rockchip released and was hacked together by them.

 

I know Sebastian and colleagues having been chipping away at Collabora ....

 

https://gitlab.collabora.com/hardware-enablement/rockchip-3588/notes-for-rockchip-3588/-/blob/main/mainline-status.md

 

I like the idea of using btop, but from what I can see it needs to be compiled, so will try that later. 

 

I recall when I had a Rock64 that, when the kernel enabled all the hardware features, the power usage did increase.  As my install is a headless server, I could probably disable the HDMI and a few other things.  

 

It's not a major deal, but it would be nice to reduce the power usage when idle.

 

Keep up the great work!

[Werner]

2 hours ago, Thewonderer said:

I believe Kernel 6.1 is the Android Kernel Rockchip released and was hacked together by them.

Actually the first bsp was that which was built around a 5.10.110 and later .160 android hacked together kernel. 6.1 is AFAIK a real linux kernel rockchip used for newer bsp.

[Jacob Olness]

Want to avoid making a new thread for this if possible but on the minimum CPU clock being higher, sounds like that's expected behavior so that's good (I actually flashed two different versions of Armbian with the 6.12 kernel as a sanity check).
I don't think it was really touched on if this is the long term plan for newer kernels or why it's set higher than the vendor kernel. I am interested both from a user perspective and from a technical one. At first the idle being so much warmer (I touched the casing and was surprised how warm to the touch it was. About 38ºC/100ºF skin temp vs 32ºC/90ºF on 6.1) was what made me worried something was wrong before looking at the available frequencies. 
 

[MaxT]

if this is the long term plan for newer kernels or why it's set higher than the vendor kernel.

I guess some insights might come from the relevant CPU opp-table evolution in mainline and rockchip mailing list on lore.kernel.org There might be some discussion.

Also you might want to create an overlay and test how it works. Just note that some existing frequency overlays do not work since use single opp-microvolt value , eg opp-microvolt = 1000000, which is target microvolt value for given frequency. But at least in some opp-tables triplets approach is used eg. opp-microvolt = 1000000 1000000 1000000 ; which are target, min and max.

Per bindings, both approaches seem legit, but they cannot be used simultaneously.

[Jacob Olness]

That's a great call. 
I am gonna do some digging there to find out. 

I just got this board so it is currently doing nothing of import so if I blow something up, no biggie. Just wanted to have a more solid idea of what I'm going to run and such before moving to the NVMe (harder to flash than an SD card haha). 

Appreciate the response!

[MaxT]

You might also be interested in tkaiser (who is sbc-bench author) thoughts:

https://github.com/ThomasKaiser/Knowledge

Not sure if there is smth specific re rk35xx frequencies, temps and energy, but definitely there is some interesting stuff re ARM overall and rk3399 in particular, which might be helpful.

[Jacob Olness]

@MaxT Sweet! I also have a rockpro64 so info about the 3399 is interesting too.