Did Linux Power Consumption Improve At All This Year?

[phoronix]

Phoronix: Did Linux Power Consumption Improve At All This Year?

As part of our end-of-year testing, a Phoronix reader had inquired about whether Linux made any strides in 2015 for improving power efficiency or extending battery life for any broad number of mobile Linux systems...

http://www.phoronix.com/scan.php?page=news_item&px=Linux-Power-Use-2015

[saski]

I can confirm this. Power Consumption on my lenovo T440p didn't change much since kernel 3.13. Especially idle comsumption is still 5-6 Watts above what Windows 10 can do. Package states above pc3 are almost never entered according to powertop. I doubt that this is gonna change for mobile HSW ever
I'm keen how newer BDW or SKL chips handle pc states when idle. Maybe someone can post his/her meassurements with powertop?

[pszilard]

I had mixed results on an MBA 6,2 (HSW), the improvements came mostly from issues related to some devices not sipping power like crazy anymore e.g. the card reader up until 15.04 was pulling serious power even when idle and some stupid interrupts were keeping the CPU from going into any low-power state.

I have not done thorough power state analysis, I could certainly do that. Is there some some standard-ish scriptable way to collect data that allows comparisons? Does the Phoronix suite allow collecting more than just power consumption?

[Michael]

I had mixed results on an MBA 6,2 (HSW), the improvements came mostly from issues related to some devices not sipping power like crazy anymore e.g. the card reader up until 15.04 was pulling serious power even when idle and some stupid interrupts were keeping the CPU from going into any low-power state.

I have not done thorough power state analysis, I could certainly do that. Is there some some standard-ish scriptable way to collect data that allows comparisons? Does the Phoronix suite allow collecting more than just power consumption?

If you set PERFORMANCE_PER_WATT=1 environment variable, it will also calculate performance-per-Watt for each test based upon what it recorded for power consumption. Or if you want other non-power metrics, set MONITOR=all environment variable.

[sarfarazahmad]

Isnt the title a little misleading "Did Linux Power Consumption Improve At All This Year?" i don't think power consumption was supposed to improve, efficiency was. I could be wrong though. English isnt my mothertongue. Just a technicality i guess

[sdack]

Well, I could upgrade my hardware this year and it lead to a lower power consumption. So Linux did let me reduce the power only did this not directly happen through a power saving option, but simply by supporting my hardware.

I have yet to try this new kernel option, which changes the scheduler algorithm to use less power. The default is to run the load across all cores and if I understand it correctly does the new kernel option implement a single batch queue, which makes it more power friendly.

Has this been looked at yet by Phoronix? Chances are the option is not enabled by the distribution's standard kernel.

Edit:
The option is called CONFIG_WQ_POWER_EFFICIENT_DEFAULT.

[Sonadow]

I can confirm this. Power Consumption on my lenovo T440p didn't change much since kernel 3.13. Especially idle comsumption is still 5-6 Watts above what Windows 10 can do. Package states above pc3 are almost never entered according to powertop. I doubt that this is gonna change for mobile HSW ever
I'm keen how newer BDW or SKL chips handle pc states when idle. Maybe someone can post his/her meassurements with powertop?

Just bought a Skylake notebook 2 weeks ago and OOTB it burns hotter under Linux (Fedora 23) than on Windows 10. the intel_pstate driver recognizes all the available scaling frequencies for the processor but stubbornly refuses to run at any other frequency besides the maximum (3.1GHz) even when on the powersave governor. While Windows 10 actually knows to downthrottle the processor to its lowest possible frequency (400MHz) when the system is idling or under very light load.

[caligula]

Do the kernel configurations affect power consumption much? For example there's that tickless option and different frequencies to choose from. For ACPI & block devices, there are some powersave options that reduce performance. For filesystems and sockets you have some memory mapped configs. What if you disable large parts of stuff in kernel? I've managed to reduce the size of the kernel to around 3,0-3,3 megs with all the functionality I need. I'd guess that a smaller kernel has less code to run when idle.

[Michael]

Isnt the title a little misleading "Did Linux Power Consumption Improve At All This Year?" i don't think power consumption was supposed to improve, efficiency was. I could be wrong though. English isnt my mothertongue. Just a technicality i guess

Many readers still report Linux consumes more power than Windows, so it's just not about efficiency.

[milkylainen]

Everybody keeps on talking about the desktop experience including power consumption. It is and has always been mostly shit. I have run Linux and built just about every kernel release since early 2.0.x. Everything that has anything to do with desktop experience is just.. horrible. Distributions keep changing their desktop layout and "ideology" every time a new release comes out. All Windows and OSX-releases have far more in common regarding user experience than any coherent Linux distribution ever had. Graphics drivers are still shit, with binary Nvidia beeing the only half usable experience. X-applications keep on hogging CPU-time although they are idling. Configuring anything and installing any package or driver is different from every distribution and keeps on getting reinvented (rpm, yum, dnf etc etc.). Configuring resolution, fonts, printers, sound, even the fricking mouse is a chore. NONE of this is appreciated by the average user.

For the average OEM there is absolutely 0 incentive of tuning anything for Linux. They try to adhere to standards, tune it for Windows and if it sort of works for Linux, then that's just fine. They know that they cannot keep up with 200 different rpm-packages of everything, every different Linux kernel gettting a reinvention of an ACPI-interpreter every now and then.

So we still get broken ACPI-tables, half assed PCIe-ASPM and other crap that Windows seem to swallow (atleast with vendors drivers).
And it probably is going to continue to be like that until Linux and all distributions stop acting like a moving Desktop target.
The server space just don't care as much about classical desktop attributes.