Need a Way to Disable Intel Turbo Boost on ACPI Events

Fastuous · 2015-07-23 03:14:26

New to the forums so I hope this is in the right place.

I have an Asus UX305 with an Intel Core M-5Y10c processor. I've been using all the default cpu drivers (intel_pstate with powersave governor). I just finished installing arch and started going through the wiki on power management for laptops.

It seemed like laptop-mode-tools had everything I needed. The intel_pstate module successfully enabled/disabled intel turbo boost on battery/AC. However I keep experiencing a weird behavior, any time the laptop resumes from suspend (regardless of lid state) turbo boost gets enabled again. I checked /proc/sys/vm/laptop_mode to make sure that laptop_mode was still running, and sure enough it was. The only way I could get turbo boost to be disabled again was to either reboot, or to plug/unplug the power cable to trigger the on-battery mode again. I tried restarting the laptop-mode.service, it didn't disable turbo boost. I tried looking at the output of laptop_mode status, it correctly reports whether the laptop is on battery/AC. No idea what's going on.

I started reading up on recommended governors/frequencies for modern intel cpu's, and found some compelling information on why you should just let intel_pstate do its thing.

https://plus.google.com/+TheodoreTso/posts/2vEekAsG2QT

So for now I just have thermald installed and letting the processor run at whatever frequency it wants. But I can't help thinking that I could be getting significant energy saving if I disabled turbo boost when on battery power.

Anyone have any opinions on cpu scaling on modern intel cpus? Or ideas of how to get laptop-mode-tools to work when resuming from suspend?

Awebb · 2015-07-23 11:44:16

Before you run in circles trying to find a way to manually disable the turbo, you should do some actual performance tests. Turn it off manually and see if it is beneficial. You might end up chasing ghosts.

Fastuous · 2015-07-23 14:18:20

That's a good idea actually.

As far as actual cpu performance goes, the difference is pretty clear. I just calculated 67M digits of pi with systest and got the following results:

Without turbo boost:
800MHz clock speed
finished in 32m 41s
reached peak of 52 degrees C

With turbo boost
2000MHz clock speed
finished in 13m 48s
reached peak of 83 degrees C

How this equates to battery savings under actual day-to-day use, I'm not sure, but I'd guess it makes at least some difference whether I'm boosting or not. I'm also not sure how I would test that except just observing over a few weeks (any better ideas?).

P.S This gave me my first look at how well passive cooling works on the new core M processors, within 20 seconds of finishing the test, the temp dropped from 80 down to 50. Now as I post this I'm down to 40. No fan cooling.

Awebb · 2015-07-23 14:49:36

To me this translates to:

Without turbo boost: Your CPU needs to be awake for 32min 41s and does not have any sleep cycles during that time.
WIth turbo boost: Your CPU is done after 13m 48s and can fall asleep after that.

I'd use some tool like powertop and run four or five of those test cycles with both states. I'd also calculate small breaks for cooldown and sync the tests for idle time.

For example (with a script):

1. start powertop, start the first test without turbo boost, wait until a total of 40 min is done (~32 min + enough time to cool down), then start the next test without turbo, wait until a total of 80 min. Repeat until you feel you have collected enough data.
2. start powertop, start the first test with turbo boost, wait until a total of 40 min is done. Start test, wait until min 80. Repeat. This allows you not only to compare the energy consumption per run, but also the total energy consumption relative to the total time the device was busy. This is important, as you do not wish to minimize the consumption per task, but maximize the battery time during a normal day of work.

Make sure nothing runs in the background, you don't want your test to be polluted by sudden busy moments.

ewaller · 2015-07-23 14:59:13

In addition to what Awebb said, Consider that the display back light is a huge load. Consider your use cases as to whether the display needs to be on during the calculations.

Fastuous · 2015-07-23 18:01:43

Roger that, I usually have my brightness set pretty low, but for testing purposes I'll probably turn it off.

I've never used powertop before, but I can already tell that there's a ton of useful information in here. I'm going to skim through the user guide on their website, but is there any metric in particular I should be paying attention to? The "Frequency Stats" tab reveals a lot, but I don't know what a lot of the parameter names in the "Idle stats" tab refer to... what should I be watching for the "total battery consumption" test we're talking about here?

Thanks so much for the help so far.

Awebb · 2015-07-23 21:51:19

Last time I used powertop (pre-2.0) it had a simple power consumption display in W.

Fastuous · 2015-08-15 17:17:36

After a fair amount of personal observation, I've decided to leave turbo boost on.

However, I've discovered that my original problem is likely a result of this bug.

https://bugzilla.kernel.org/show_bug.cgi?id=66581

Guess I'll just keep on the lookout for a fix.

Arch Linux

#1 2015-07-23 03:14:26

Need a Way to Disable Intel Turbo Boost on ACPI Events

#2 2015-07-23 11:44:16

Re: Need a Way to Disable Intel Turbo Boost on ACPI Events

#3 2015-07-23 14:18:20

Re: Need a Way to Disable Intel Turbo Boost on ACPI Events

#4 2015-07-23 14:49:36

Re: Need a Way to Disable Intel Turbo Boost on ACPI Events

#5 2015-07-23 14:59:13

Re: Need a Way to Disable Intel Turbo Boost on ACPI Events

#6 2015-07-23 18:01:43

Re: Need a Way to Disable Intel Turbo Boost on ACPI Events

#7 2015-07-23 21:51:19

Re: Need a Way to Disable Intel Turbo Boost on ACPI Events

#8 2015-08-15 17:17:36

Re: Need a Way to Disable Intel Turbo Boost on ACPI Events

Board footer