Benchmarks - Is PowerTOP Tuning Worthwhile For Modern AMD Linux Laptops?

[Adarion]

I have partially similar experiences. I do wall measures with a comparatively cheap Watt-meter. It depends, of course, on the machine, and also on the number of things "bad" and "good", but just from a momentary snapshot of power consumption I can hardly detect much change. Some devices, e.g. my Geode LX would see some 0.3 ... 0.5 savings iirc. some years ago, wich is fine for some low power device. But on desktops it varies between -0.5 (sic!) and 1.5 W for idle consumptions between 16 and 60 W. (Yes, once it got worse, even minutes later with no real programs running that could cause CPU spikes.) But then, on some systems I use some kind of powertop --auto-tune or the echo "..." > /sys/... in scripts already.
A few devices, however, even dislike it. E.g. a Logitech mouse that simply switches off every 5 seconds, unless you trigger a click.
I still have powertop on all my systems. But there are other things that will gain more savings, but those are usually driver stuff in the kernel (PCIE ASPM was a good step back in the days, and all PM for Radeon GPUs; still not sure if BACO works correctly.)

[syrjala]

I was a bit curious about the claim that the kernel enables everything needed by default. So I removed my standard udev rules and ran a few idle tests. Unfortunately the results weren't exactly impressive. The kernel is still far too conservative in enabling this stuff automagically.

With a bit of tuning I can get this IVB X1 Carbon down to ~3W idle power consumption (display and wifi off). With the defaults I get ~4.5W instead.
OK, half of that 1.5W difference seems to be SATA ALPM which the BIOS doesn't enable on this machine, so that could be a non-issue on some other machine.
The other half seems to be just all the useless PCI/USB devices sucking away at the battery.

So I think I'll put my proven udev rules right back I basically drop some variant of these onto every machine that crosses my path.

#GOTO="pmend"
ACTION=="add|change", SUBSYSTEM=="pci|usb", TEST=="power/control", ATTR{power/control}="auto"
ACTION=="add|change", SUBSYSTEM=="pci|usb", TEST=="power/wakeup", ATTR{power/wakeup}="disabled"
ACTION=="add|change", SUBSYSTEM=="usb", TEST=="product", TEST=="power/autosuspend_delay_ms", ATTR{product}=="*[kK]eyboard*", ATTR{power/autosuspend_delay_ms}="30000"
ACTION=="add|change", SUBSYSTEM=="usb", TEST=="product", TEST=="power/autosuspend_delay_ms", ATTR{product}=="*[mM]ouse*", ATTR{power/autosuspend_delay_ms}="30000"
#ACTION=="add|change", SUBSYSTEM=="scsi_host", TEST=="link_power_management_policy", ATTR{link_power_management_policy}="min_power"
ACTION=="add|change", SUBSYSTEM=="scsi_host", TEST=="link_power_management_policy", ATTR{link_power_management_policy}="med_power_with _dipm"
ACTION=="add|change", SUBSYSTEM=="net", ENV{DEVTYPE}=="wlan", RUN+="/usr/sbin/iw dev $env{INTERFACE} set power_save on"
LABEL="pmend"

[sireangelus]

Then you either have buggy firmware, buggy hardware or it is a kernel bug. The kernel nowadays enables all the important things like PCIe ASPM, USB autosuspend, NVME APST and so on by default, if it deems it safe to do so. Cooperation by the firmware (ACPI) is also required to some degree. Sure you can force on some specific power saving setting, but usually there is a good reason if the kernel doesn't enable it by default.

A typical example would be PCIe ASPM: on some buggy ACPI implementations, ASPM is configured to an "always on" configuration. ASPM works fine, but the firmware simply doesn't configure a suitable policy. If you have such a probkem, file an issue with the vendor, or just make sure to buy Linux compatible hardware beforehand.

Personally I don't have much tolerance for these kinds of problems anymore. If hardware has obvious firmware bugs or Linux compatibility issues, I send it back.

whatever answer is, declaring powertop useless on AMD laptop basing on a sample of a size of 2 (you and michael) seems excessive and approximative.

[t.s.]

I have unresponsive Logitech wireless mouse-keyboard combo when tuning usb power usage using powertop. Using at least (minimum) 4 models of Logitech wireless mouse-keyboard combo with the same effect. Until now, never using auto-tune because of that.

[syrjala]

I have unresponsive Logitech wireless mouse-keyboard combo when tuning usb power usage using powertop. Using at least (minimum) 4 models of Logitech wireless mouse-keyboard combo with the same effect. Until now, never using auto-tune because of that.

If you look through my udev rules I posted earlier I set a long autosuspend delay for USB keyboard/mice. That avoids them going to sleep while in active use them, but still allows them to power down when not in use. I found 30 seconds to be an decent value for me.

[alcalde]

Idle power is important - unless you're background-folding proteins or mining coin, a typical workstation is going to be idle most of the time.

It's a laptop. Why would you have it on, on battery power, and leave it sitting idle? Turn it off and you'll achieve a 100% power savings.

[yump]

It's a laptop. Why would you have it on, on battery power, and leave it sitting idle? Turn it off and you'll achieve a 100% power savings.

1. Whenever you stop scrolling to read? Whenever you stop typing to think? That's idle. When royce said most of the time, they meant most of the time.

2. S3 sleep is 9000x better than turning your laptop off. 99% of the power savings, and it's completely non-destructive for most workflows that don't rely on persistent network connections (for ssh, use mosh), so you can set it to automatic on an idle timer.

[royce]

1. Whenever you stop scrolling to read? Whenever you stop typing to think? That's idle. When royce said most of the time, they meant most of the time.

That is exactly what I meant 👍