Which one is better at 150% zoom? Mutter or plasma?
Measure it and see. I have no idea.
Uh, sure, that was what I was aiming at.
My question was if the increase in power draw was independent of the compositor, in which case the generalisation would fit, or if it differs between compositors and we need to measure.
It will differ from compositor to compositor quite significantly I would imagine. Wayland compositors are basically like the X11 Server, bit of the Client and the Window Manager all wrapped up into one. How they implement things differs significantly. Weston is the reference wayland compositor ; wlroots is a minimalist implementation (of which sway makes use of ) and kwin_plasma differs significantly to gnomes mutter. Things like the X11 bridge in kwin_wayland are matched with the compositor vs mutter which traditionally has called a helper. Especially for legacy X11 apps - this bridge service is going to be likely the source of a lot potential optimization or not. kwin_waylands much better at dealing with it than mutter, although mutter is catching up.
FYI folks PPD 0.21 was just released.
This release should help with battery life on F13 and F16 AMD by more efficient tuning in battery mode.
Ping your distros to pull it in and try it out.
14 Likes
Seems it now automatically switches profiles on battery/mains use. More info in Phoronix too:
1 Like
I got even lower battery wattage using TLP this time
-
Main: AMD 7840U - 61Wh battery
SSD: SK Hynix Platinum P41 2TB
RAM: Framework DDR5-5600 - 1 X 32GB(likely badge engineered ADATA DDR5-5600 SO-DIMM)
Expansion cards: 2 X USB-C, 2 X USB-A -
OS: Arch Linux 6.8.4
DE: LXQt 1.4.0
Power management: TLP 1.6.1 -
0% brightness, no keyboard backlight
Wifi connected, bluetooth off.
As there seems to be a lot to sort through here, we recommending to just use powertop --auto-tune for now?
powertop --auto-tune is a big hammer that has potentially negative implications especifically for USB devices. I’ve personally observed on F16 this makes the USB keyboard feel “laggy” when it goes into auto suspend.
Using PPD 0.21 on balanced you should get a very good experience.
2 Likes
Interesting, I haven’t had that issue on the 7840U FW13 running Fedora 40 gnome. I’ll keep that in mind, though.
Is it recommended to just use the in-distro PPD for that then?
Yes F40 has PPD 0.21 already.
If you see other MEASURABLE improvements from powertop --auto-tune I would like to know.
It’s important to work out individual knobs that make a difference, not just ones that say they’re in a “bad” state to start.
1 Like
I haven’t done any measurements myself for now, but I will keep that in mind, thank you.
I discovered that in tlp.conf if you set
CPU_ENERGY_PERF_POLICY_ON_BAT=balance_powertoCPU_ENERGY_PERF_POLICY_ON_BAT=power`
It could significantly increase efficiency on light load without sacrificing performance. This is different than “power saver” on PPD as this caps the tdp so that the CPU will never exceed 85C. “power” on TLP significantly reduces the CPU frequency under low utilization(below 1.4GHz), while still allows max power usage under full loads (3.6GHz all core all the way to 100C and up to 4.6GHz single core).
This could be very useful at web browsing or video streaming
EDIT: apparently both TLP and PDD does this, but the CPU frequency “ramp-up” times are slightly different on powersave. This still needs more testing
I wonder what config did you use in TLP for 7840u. Did you follow the instruction for intel CPU and apply exactly on AMD CPU?
This strategy was remarkable years ago on some of my older laptops, so I included it in my testing regimen. It did not have a significant impact, perhaps unsurprisingly in retrospect with advancements like NVMe drives leaving little on the table for powersaving.
Not only that; but myself and others have made an effort to setup policies in the kernel and udev to set the things that really matter.
4 Likes
for some reason, PPD 0.21 drastically increased suspend power consumption, the problem is fixed by downgrading PPD of using TLP
That doesn’t make sense, can you please get an s2idle report with both versions and a cycle that is at least 5 minutes programmed for it? It will quantify with measurements.