Can powertop power values be trusted?

Framework Laptop 13 Linux
1

I’m trying to lower the power usage on my Framework 13 AMD 8-core 2.8k display laptop.
So I have been investigating using powertop and it shows for example that the WiFi card would be using about 7-6W of power. The total power draw of the battery is reported as around 10W at idle with no applications loaded, so that’s quite substantial.
Are these number real? Is the WiFi card really using that much power?
Beginning of overview page:

The battery reports a discharge rate of 9.28 W
The energy consumed was 187 J
The estimated remaining time is 6 hours, 54 minutes

Summary: 351,8 wakeups/second,  0,0 GPU ops/seconds, 0,0 VFS ops/sec and 3,9% CPU use

Power est.              Usage       Events/s    Category       Description
  6.88 W      4,6 pkts/s                Device         Network interface: wlp1s0 (mt7921e)
  156 mW      6,9 ms/s     121,6        Interrupt      [117] amdgpu
 78.9 mW      1,3 ms/s      61,6        Timer          tick_nohz_highres_handler
 20.2 mW      1,1 ms/s      15,8        Interrupt      [7] sched(softirq)

Beginning of device status page:

The battery reports a discharge rate of 9.42 W
The energy consumed was 191 J
System baseline power is estimated at 7.59 W

Power est.    Usage     Device name
  6.87 W      4,3 pkts/s  Network interface: wlp1s0 (mt7921e)
  710 mW      5,5%        CPU misc
 9.58 mW      5,5%        CPU core
    0 mW     16,5%        Display backlight
1 Like
2

From what I recall reading, no, the values can not be trusted at all.
Someone correct me if I’m remembering wrong.

2 Likes
3

IIRC, the reported total power draw is accurate, but it’ll be misattributed. Battery reporting rate can be verified with upower

upower -i /org/freedesktop/UPower/devices/battery_BAT1 | grep energy-rate

and powerstat -R to check the CPU package.

Also happy to hear about any reliable tools for
power measurement on AMD

3 Likes
4

what I found is the measurements are more accurate when your only using battery, don’t use powertop when using your AC power adapter

freaking annoying but yeah

1 Like
5

The battery draw is as accurate as the the battery reports it, the rest is somewhat educated guesses.

If you run powertop calibrate the guesses get more accurate but with all the power saving tricks these days it keeps getting harder to get accurate attribution. Sttuff like the draw of the display can get quite accurate after calibration, at least for lcds with fixed backlights.

1 Like
6

Yeah, I forgot to mention that part but that’s what I was referring to

1 Like
7

Thanks for the replies! Very insightful
The lowest I have gotten the battery drain was ~7.86 W. That was with all expansion cards removed and using single user mode (init 1), no X-server loaded, all cores (threads) except 4 disabled, display refresh rate at 30 Hz, power saver profile, display at about 10%. Kernel 6.8.0-45-generic.
A bit high I think… The bottom of laptop gets mildly warn.

8

That does sound high.

I’m comfortably getting ~4W at idle in GNOME with 2.8k display at 40%.

I’d recommend latest kernel 6.11.1 and ppd 0.23 as a start.

9

That sounds pretty high, I managed to get into the high 2s with sway running display on, min brigtness, wifi connected and bt on. Doubt the 2.8k screen draws that much more than the 2k one.

In more realistic use I get about 4ish idle with a few firefox windows and a bunch of terminals open but not doing much. Video playback got a lot better recently but still draws quite a bit of power and the touchpad also still seems to cause power use spikes but it has gotten pretty good lately.

1 Like
10

There was Framework 13/16 AMD Linux battery life improvement testing a while ago, looks like the patch stalled though.

11

I’m running the latest Linux Mint 22 (based on Ubuntu 24.04). I tried to install 6.11 from mainline, but either it failed booting because of secure boot or it failed to build if I tried to use it with with a custom MOK. Something about fwts-efi-runtime-dkms failed to build.
So I guess I’m stuck on 6.8 for now.

12

Just uninstall fwts.

13

Have not kept up with that though the spikes have certainly seemed to be a lot less severe with recent updates. Still a lot worse than using a bt mouse but nowhere near as extreme as they used to be.

14

That’s definitely too high. Something doesnt seem right. Are you sure you are idling? Nothing puking out errors into the system log? What do you get booting fedora 40 or ubuntu 24.04 liv with same setup?

15

I have now tested a bit. I got Kernel 6.11.0 working with secure boot, however the power consumption figures were not lower.
My current Mint 22 system with 6.11.0: Battery discharge rate lowest at about 8.55 W, average CPU consumption (powerstat): 1.40 W.
With Fedora 40 Live CD: Battery discharge rate 7.63 W, CPU consumption 0.75 W.
With Ubuntu 24.04.01: Battery discharge rate 5.68 W, CPU consumption 0.56 W.

These numbers were at a stable idle state. The Mint 22 numbers were from my installation so that contains some additional installed applications and services so I expect a bit higher numbers as compared to the Live CDs.
The Ubuntu results were a bit lower than the others, interesting! Not sure why…

16

You could install system76-power and see if that makes a difference. For me switching to the battery saving profile it is currently using 3.3W and in balanced consumption was around 5W.

17

I tried system76-power but it didn’t change anything. Still the same power draw with their battery profile.
I guess it’s just Linux MInt that is not adapted for the Framework, and not much that can be done about it at the moment.