Announcement

Collapse
No announcement yet.

Radeon DPM Is Fantastic For Power Use, Thermal Performance

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

  • #16
    Originally posted by curaga View Post
    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.
    WebGL support: Chrome, Firefox, Safari, and IE starting with v 11 and Opera is working on it.
    WebGL had security issues which were fixed including the WebGL specs which led Microsoft to drop the security accusations against WebGL and embrace it. It can waste a ton of power if the programmer is stupid - same is true for any other code in any other language.
    Firefox has one more solution for rogue JS scripts (WebGL is powered by JS too): it stops JS from executing on all tabs except the current one.

    Comment


    • #17
      Aside from the graph colors, I really enjoyed this article, particularly this bit on page 3:
      The margin is quite small for how the Radeon HD 6870 performance was impacted with DPM enabled, plus the performance was already 200+ FPS, so I'm really not too worried. It's possible the performance could be negatively impacted if the Radeon DPM code is too conservative with its frequency scaling and takes too long to ramp up the clock frequencies, but that's hard to tell as there isn't any convenient sysfs/debugfs interface right now for being able to poll the core/memory frequencies in a real-time manner.
      I'll take your word for it about the sysfs/debugfs part, but at least I appreciate the analysis instead of the usual description of the graph results. More of this please! That's exactly what I expect from serious benchmarks.

      Note: the text under the first graph on page 5 only reads "Unvanquished", is that a place-holder you forgot to edit?

      Comment


      • #18
        Originally posted by mark45 View Post
        WebGL support: Chrome, Firefox, Safari, and IE starting with v 11 and Opera is working on it.
        That doesn't answer the question: browsers X, Y and Z may support it, but the question was about installed base. Are those supporting versions actually being used.

        Comment


        • #19
          Originally posted by curaga View Post
          That doesn't answer the question: browsers X, Y and Z may support it, but the question was about installed base. Are those supporting versions actually being used.
          Were you born yesterday? Don't know how things work? Never heard about trade-offs and fallback solutions?

          Comment


          • #20
            Actually this benchmark has a quite limited value. I know using one desktop base and just swapping cards was quite fast and convenient, but power management matters for notebook users.
            This article answers the 2 questions how new DPM affects 3D performance and what is power usage/temps for near maximum load. What it lacks is answer for "how dpm is affecting battery life" and "how effective it is at downlcocking/disabling parts of gpu when they are not in use" questions.
            Good PM code benchmark should report battery time , power consumption histogram and average power consumption under different loads (idle, light desktop usage, playing videos etc) on various AMD GPUs/APUs equipped lappys compared to Catalysts and FOSS driver with DPM disabled.

            Comment


            • #21
              Originally posted by Xeno View Post
              Actually this benchmark has a quite limited value. I know using one desktop base and just swapping cards was quite fast and convenient, but power management matters for notebook users.
              This article answers the 2 questions how new DPM affects 3D performance and what is power usage/temps for near maximum load. What it lacks is answer for "how dpm is affecting battery life" and "how effective it is at downlcocking/disabling parts of gpu when they are not in use" questions.
              Good PM code benchmark should report battery time , power consumption histogram and average power consumption under different loads (idle, light desktop usage, playing videos etc) on various AMD GPUs/APUs equipped lappys compared to Catalysts and FOSS driver with DPM disabled.
              I don't even have any laptops with AMD GPUs aside from some old R300 and R500 era ThinkPads...
              Michael Larabel
              http://www.michaellarabel.com/

              Comment


              • #22
                Aside form desktop/notebook issue, is it really pm benchmark? Comparing powersaving capabilities of drivers in 3D gaming scenario is such an odd case. When you run games you want performance first.. It's nice to see it can save some watts under heavy load, but what does it say about impovement in power saving? It won't tell you if dpm does better job than "low" porfile from old code when system is idle (both watts/h and temps). How much it's better (performace/watts/h, temp) in 2d/video compared to profles and old dynpm.
                That would be somethin closer to the cases when user wants to save power.

                Comment


                • #23
                  Originally posted by mark45 View Post
                  Were you born yesterday? Don't know how things work? Never heard about trade-offs and fallback solutions?
                  Obviously it's your time to lose if you want to serve the 1% while the majority gets the fallback. I just find curious how you advocate such a minority tech.

                  Comment


                  • #24
                    Originally posted by Michael View Post
                    It comes down to not having a good automated color picking algorithm.

                    Well theres your problem. Set your colors manually for readability. You still rely far too much on automation.

                    Comment


                    • #25
                      Originally posted by Michael View Post
                      It comes down to not having a good automated color picking algorithm.
                      http://www.dtic.mil/dtic/tr/fulltext/u2/a272919.pdf

                      page 28 describes the algorithm

                      You're welcome



                      To the rest: I'd love to know what's going on with the 6770. That seems to be, by far, the most dynamic of the bunch, while the oldest card, 4770 I think, was the least dynamic.
                      That's suggests differences in hardware as the cause since the older cards are generally better opitmized in driver.
                      Last edited by liam; 07-31-2013, 04:35 PM.

                      Comment


                      • #26
                        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.

                        2. The line graphs were pretty much unreadable. You had black, blue, and then 4 separate shades of red that all ran together.
                        On 1.
                        Only when you used a static "low" profile, the other profiles should benefit.
                        On 2.
                        Good to see I am not the only one. I am even quite good at keeping colours apart but that was a graph that was hard to interprete.

                        Comment


                        • #27
                          What language are you using for your automated color picker? Is it posted where we can take a look at it and try to fix it for you?

                          Comment


                          • #28
                            Originally posted by ua=42 View Post
                            What language are you using for your automated color picker? Is it posted where we can take a look at it and try to fix it for you?
                            PHP. mostly within http://phorogit.com/index.php?p=phor...95fb08252c87d3
                            Michael Larabel
                            http://www.michaellarabel.com/

                            Comment


                            • #29
                              Very good power management!

                              I can also confirm power management is very good for me on 5750 and 5770 cards. In idle states GPU are about 10C colder than they were by default, thanks to lowest frequency possible. Then they quickly increase frequency on load, so there was no noticeable performance loss either. Really remarkable improvement!

                              And btw, I can monitor GPU frequencies like this:
                              Code:
                               while (true) do cat /sys/kernel/debug/dri/0/radeon_pm_info; sleep 1; done
                              And file contains something like this for 57xx GPUs:
                              Code:
                              uvd    vclk: 0 dclk: 0
                              power level 0    sclk: 15700 mclk: 30000 vddc: 1000 vddci: 1000
                              So I guess its possible to monitor GPU frequency in tests as well. Though it changes quite often.
                              Last edited by 0xBADCODE; 08-01-2013, 07:55 AM.

                              Comment


                              • #30
                                Originally posted by Michael View Post
                                It comes down to not having a good automated color picking algorithm.
                                Easiest way is to just use a predefined palette eg http://web.media.mit.edu/~wad/color/palette.html

                                Other answers including a simple python function http://stackoverflow.com/questions/4...istinct-colors

                                Comment

                                Working...
                                X