Announcement

Collapse
No announcement yet.

ACPI CPUfreq vs. Intel P-State Scaling With Linux 3.15

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • #16
    Originally posted by liam View Post
    Would do mind attaching that wattmeter to the Nvidia board so we can get some official power numbers?
    Thanks. Yes I do plan to, albeit I only have one WattsUp meter and am constantly using it on many different systems so unfortunately it takes a while. I hope to have the power numbers for the Jetson TK1 out next week.
    Michael Larabel
    http://www.michaellarabel.com/

    Comment


    • #17
      intel pstate

      Originally posted by Michael View Post
      Thanks. Yes I do plan to, albeit I only have one WattsUp meter and am constantly using it on many different systems so unfortunately it takes a while. I hope to have the power numbers for the Jetson TK1 out next week.
      intel pstate is bad, really bad with battery life is bad with temps in idle and performance state dont works well

      Comment


      • #18
        Originally posted by Michael View Post
        Thanks. Yes I do plan to, albeit I only have one WattsUp meter and am constantly using it on many different systems so unfortunately it takes a while. I hope to have the power numbers for the Jetson TK1 out next week.
        That's great. I look forward to it.

        Comment


        • #19
          Colin Ian King, kernel engineer tested P state extensively with a watts meter to find out higher consumption of pstate versus acpi cpufreq and its on his recommendation pstate was disabled for Ubuntu 14.04 https://lists.ubuntu.com/archives/ke...il/041515.html

          His experience reflects mine on Arch with i7 IVB laptop. I notice higher temps and more consumption, with TLP I get 4.5 hours on Ubuntu 14.04 as opposed to 3.5 hours with Arch with latest kernel 3.14 and pstate.

          Comment


          • #20
            Few remarks (please correct me if I get anything wrong):

            P-state has only one governor (default) which can be tweaked by 3 settings (min,max,no_turbo) via sysfs.
            P-state doesn't really change frequencies, everything is running at max frequencies with sleep-states inserted according to load.

            So in workloads:

            gaming - everything runs on max, cpu doesn't sleep unless it's forced to by high temperature
            semi-idle - everything runs on max, cpu sleeps most of the time, when there are tasks to be taken care of it wakes up and goes to sleep
            idle - cpu sleeps (apart from one core that is running scheduler and poriodic tick)

            It came to me as a big surprise when in past posts about P-state there were 'performance' and 'powersave' governors.
            <pre>
            echo 100 > /sys/devices/system/cpu/intel_pstate/max_perf_pct
            echo 26 > /sys/devices/system/cpu/intel_pstate/min_perf_pct
            echo 0 > /sys/devices/system/cpu/intel_pstate/no_turbo
            </pre>

            Comment

            Working...
            X