No announcement yet.

Linux 4.17 Offers Some Promising Power-Savings Improvements

  • Filter
  • Time
  • Show
Clear All
new posts

  • #31
    Originally posted by dsmythies View Post

    What is your normal idle power consumption? i.e. how much better is it now?
    EDIT: the savings on the idle test from this report seem to be about 20%, and for a similar processor. I'd sure like to see the idle state 0 data for this test.

    So far, the larger idle system power savings have been on bigger systems, with many 10s of CPUs. Drastic improvement in idle system power consumption for smaller systems is not an expected outcome. One also has to be careful how they define "idle", as they can be quite different between a server with minimal services running and a desktop with a GUI.

    For my older i7-2600k server (with most services disabled. i.e. really very "idle") I get 3.6% processor package power saved.
    However for the Phoronix hiemno test (Single, CPU intensive, thread) gives 1.1% increased performance and a 9.3% processor package power improvement.
    You're right: the power savings are not that significant. I was going off memory from my previous configuration thinking that it idled around 5-6W, but having back to kernel 4.16, I see that it does idle under 5W still. Unfortunately I've got issues with 4.17-rc1 so I won't retest until a working RC comes out.


    • #32
      Originally posted by schmidtbag View Post
      Wow that's some pretty good results with practically no regressions. This is some pretty exciting results. A little strange how AMD didn't seem to be affected, but I guess it's not too surprising.
      Numerous months later...

      I decided to use ukuu to run the latest kernel on my HP 14-cm0011na Ryzen 2500U laptop (which I'm delighted with as a serious lightweight developer machine at £530), and have updated Ubuntu 18.04.1 from 4.15 to 4.20.

      My comparisons using powertop are showing a 0.2 - 0.3W improvement at idle, bringing min idle power down to 7.3W, so I'm going to give 4.20 a go for a while, and see if I get any useful improvements in time spent doing development.

      It wouldn't surprise me if I find a performance improvement given the 15W TDP limit that I'll be often hitting.


      • #33
        Until we get new chipsets out, none of this is applicable due to the spectre "fixes".
        20 to 50 percent slowdowns vs these paltry power efficacy improvements?