Announcement

Collapse
No announcement yet.

Radeon DPM Is Fantastic For Power Use, Thermal Performance

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

  • Radeon DPM Is Fantastic For Power Use, Thermal Performance

    Phoronix: Radeon DPM Is Fantastic For Power Use, Thermal Performance

    One of the most exciting features of the upcoming Linux 3.11 kernel is the open-source Radeon driver's support for dynamic power management (DPM). We have already done preliminary benchmarks and found that Radeon DPM can boost the GPU's performance in cases where the boot clock speeds are slower than their rated frequencies (as in the case of AMD APUs and modern high-end GPUs). For other GPUs, Radeon DPM can lead to lower power consumption and better operating temperatures. Here's looking at the Linux Radeon DPM performance with the Linux 3.11 Git kernel.

    http://www.phoronix.com/vr.php?view=18980

  • #2
    Ok.... I have to do it

    Some places that could use improvement here, Michael.

    1. Test idle power usage. Honestly it shouldn't change much, if at all, while you are running apps. It's at the desktop that you should really see changes.

    2. The line graphs were pretty much unreadable. You had black, blue, and then 4 separate shades of red that all ran together.


    It looks to me like the 6870 isn't quite getting into the highest power state under DPM. Not only were the benchmark scores consistently a bit lower, it was by far the biggest recipient of the power efficiency gains (meaning it used far less power).
    Last edited by smitty3268; 07-30-2013, 10:13 PM.

    Comment


    • #3
      Originally posted by smitty3268 View Post
      Some places that could use improvement here, Michael.

      1. Test idle power usage. Honestly it shouldn't change much, if at all, while you are running apps. It's at the desktop that you should really see changes.
      It depends on the defaults. If it's defaulted at low clocks, DPM should increase power consumption but come with increased performance. So, I think both benchmarks are needed to judge.

      2. The line graphs were pretty much unreadable. You had black, blue, and then 4 separate shades of red that all ran together.
      Completely agree.

      Comment


      • #4
        Not to author- your graphs are so confusing I couldn't make heads or tails out of them.

        You are using virtually the same colours for multiple graphs.

        Couldn't you use stipples and/or separate graphs - one for each card ?

        Comment


        • #5
          Originally posted by smitty3268 View Post
          2. The line graphs were pretty much unreadable. You had black, blue, and then 4 separate shades of red that all ran together.
          It comes down to not having a good automated color picking algorithm.
          Michael Larabel
          http://www.michaellarabel.com/

          Comment


          • #6
            4 shades of red....

            Comment


            • #7
              Originally posted by Michael View Post
              It comes down to not having a good automated color picking algorithm.
              Just reduce the set from which you chose.
              This sounds like a good choice:
              http://www.w3.org/TR/REC-html40/types.html#h-6.5

              Comment


              • #8
                Originally posted by Michael View Post
                It comes down to not having a good automated color picking algorithm.
                Honestly at this point a random number generator could have picked better colors.

                But yeah, this is an old topic.

                Comment


                • #9
                  How does it all compare to Catalyst drivers?

                  Comment


                  • #10
                    Michael why don't you use a constellation like in telecommunications to pick up colors? It should be quite easy to implement and very, very effective.
                    ## VGA ##
                    AMD: X1950XTX, HD3870, HD5870
                    Intel: GMA45, HD3000 (Core i5 2500K)

                    Comment


                    • #11
                      Great news and great benchmarks Michael. I think you should pick colours that seperate easier from each other. It was impossible to read some graphs with those colours.

                      Comment


                      • #12
                        CPU governors where also reported as having some (sometimes dramatical) influence on Mesa/Gallium results.

                        Michael could You test it too? (radeon.dpm and ondemand/performance/conservative governors)

                        Comment


                        • #13
                          I have a AMD notebook, so this benchmark is very interesting for me. The battery will last more time. :-)


                          Originally posted by Michael View Post
                          It comes down to not having a good automated color picking algorithm.
                          Well, I think a good way would be pre-select the colour like:
                          - one line: black
                          - two lines: black and red
                          - three lines: black, red and green
                          - four lines: black, red, green and dark blue
                          - five lines: black, red, green, dark blue and dark yellow
                          - six lines: black, red, green, dark blue, dark yellow and light blue
                          - seven lines: black, red, green, dark blue, dark yellow, light blue and pink
                          etc
                          Of course less lines would be the better contrast, but it's just a suggestion.

                          Obs.: Michael, it would be good if you do some APU benchmarks using DPM

                          Comment


                          • #14
                            It would be even better if Michael moved from "server creates base64 image, dumb client displays it" to interactive stuff like on other professional web sites like e.g. net applications or gs.statcounter.com.
                            I'm not a flash programmer but I'm doing some WebGL stuff (which IE will also support starting with IE11) and if Michael would like I'd help develop the interactive client side in WebGL for free, but this would require quite some changes on the server side too which he's unlikely to be willing to commit to since it's a one man show.

                            Comment


                            • #15
                              How much of the client base supports webGL? Is it over 10% yet?

                              Personally, I would never enable it even if my browser did support it - it's a huge security hole, while also allowing your browser to waste a ton of power mining bitcoins or showing some really shiny ad.

                              Comment

                              Working...
                              X