Announcement

Collapse
No announcement yet.

Radeon Power Management Gets More Fixes For 3.11

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

  • #16
    Originally posted by verde View Post
    I am not that excited (with the DPM) about recent AMD GPUs as long as they have the FGLRX option for modern performance and power management.
    I disagree. The power management on Radeon is apparently very comparable to that of FGLRX now. The performance is still not acceptable for "real" gamers, but for many people it's more than enough.

    Also, many laptop users needed to install FGLRX (and therefore sacrificing stability) simply because of heating and fan issues. The DPM seems to solve that. And I doubt many laptop/netbook users are going to be doing much gaming.

    Comment


    • #17
      Originally posted by Rexilion View Post
      Why is Windows not suffering from these kind of issue's?
      It's mostly there for debugging since we introduced two new independent features at the same time. The only board where I've seen issues with ASPM so far was an internal engineering sample board. For windows we generally work directly with the OEM at a platform level so everything is validated.

      Comment


      • #18
        Originally posted by agd5f View Post
        It's mostly there for debugging since we introduced two new independent features at the same time. The only board where I've seen issues with ASPM so far was an internal engineering sample board. For windows we generally work directly with the OEM at a platform level so everything is validated.
        Hey, Alex, complete random off the wall question.... Is cloned outputs harder on an AMD GPU than single output, or no difference? I've got a laptop running Win7 that I use for netflix, gaming media center, etc. The laptop's fscked up so its basically just a shell, cuz I've got an HDMI cable running to a TV and a usb mouse and keyboard that I use from afar. The outputs (laptop screen and hdmi out) are cloned, not extended. Does the GPU care? Like is it rendering the scene twice, or is it rendering once and just submitted the buffer to two outputs?

        Just wondering if it matters to decide whether or not I should go into Catalyst Control and turn off the LVDS output (for the sake of frame-rate and power consumption of the GPU), or just leave it on.

        Comment


        • #19
          Originally posted by Newfie View Post
          I disagree. The power management on Radeon is apparently very comparable to that of FGLRX now. The performance is still not acceptable for "real" gamers, but for many people it's more than enough.

          Also, many laptop users needed to install FGLRX (and therefore sacrificing stability) simply because of heating and fan issues. The DPM seems to solve that. And I doubt many laptop/netbook users are going to be doing much gaming.
          To my taste, AMD did a great improvement in both areas lately. They both got a decent power management and then MESA 9.2 also faster in many workloads. Also UVD decode support. A really noteworthy amount of improvements. So in fact everyone except most hardcore gamers would be fine with opensource drivers and generally these are far more stable than fglrx and never cause prob's with install/deinstall, system updates, etc, etc. Well done!

          Comment


          • #20
            Originally posted by Ericg View Post
            Hey, Alex, complete random off the wall question.... Is cloned outputs harder on an AMD GPU than single output, or no difference? I've got a laptop running Win7 that I use for netflix, gaming media center, etc. The laptop's fscked up so its basically just a shell, cuz I've got an HDMI cable running to a TV and a usb mouse and keyboard that I use from afar. The outputs (laptop screen and hdmi out) are cloned, not extended. Does the GPU care? Like is it rendering the scene twice, or is it rendering once and just submitted the buffer to two outputs?

            Just wondering if it matters to decide whether or not I should go into Catalyst Control and turn off the LVDS output (for the sake of frame-rate and power consumption of the GPU), or just leave it on.
            Display controllers are like pointers. They point to a location in vram and read what's there, so there's no additional rendering required for cloned displays, but each additional display controller uses power and the more display controllers that are active, the more contention there is for memory bandwidth in the GPU. It's pretty much the same situation regardless of the vendor.
            Last edited by agd5f; 07-21-2013, 09:27 AM.

            Comment


            • #21
              I just compiled the RC-2 and it is still not working here on a RV635/HD3650 even with the radeon.aspm=0 parameter. Systematically hangs at boot with dpm enabled.
              Last edited by Scias; 07-21-2013, 11:46 PM.

              Comment


              • #22
                Originally posted by Scias View Post
                I just compiled the RC-2 and it is still not working here on a RV635/HD3650 even with the radeon.aspm=0 parameter. Systematically hangs at boot with dpm enabled.
                Thousand times this. A white snowy like screen appears and it gets whiter and whiter. The system never loads.
                Could you fix it adg5?

                Or do you have a fix available that we could test?

                Comment


                • #23
                  3.11 RC2 here with radeon.dpm=1: my E-450 laptop hangs at suspending / resuming (sometimes at suspending, sometimes at resuming). Please, check the resume/suspend code.

                  However, I'll leave 3.11 RC2 installed, since it gives me the same battery life of Catalyst, and I also got that bug with Catalyst (it's a bug in the original AMD code). If you know how can I get logs of suspending / resuming, please, tell me.

                  BTW, all Fedora users, please, test again: there is a Mesa-9.2-git_snapshot_from_yesterday package set in Koji; that package includes the Radeon 2D performance boosts published on Phoronix.

                  Comment


                  • #24
                    I tried the latest kernel from Alex's fork quickly for the most simplest things for my Intel 4000 + HD 7970M again.

                    Booting with dpm=1 works okay.
                    rmmod radeon worked.
                    modprobe radeon dpm=1 took several seconds and something in the UVD initialization went wrong. Doesn't really matter, it worked and I won't use the UVD anyway.
                    rmmod radeon got me a complete system lockup. In the past rmmod radeon did always lockup the first time.
                    In previous versions I tried vgaswitcheroo and always got a compete lockup too.
                    Come on, AMD that is the top of the line mobile gpu from you. I understand if radeonsi rendering is not good yet. But system lockup when trying to power save or reloading the driver? Come on!

                    On the other hand I get complete lockups too on the intel driver on the latest stable kernel too. It's not even hard to reproduce. Play half life 2 deatmatch for a few hours and you have a very good chance that your system locks up.

                    I want my 2008 linux back that didn't lock up all the time. :/

                    Comment


                    • #25
                      Originally posted by Alejandro Nova View Post
                      3.11 RC2 here with radeon.dpm=1: my E-450 laptop hangs at suspending / resuming (sometimes at suspending, sometimes at resuming). Please, check the resume/suspend code.

                      However, I'll leave 3.11 RC2 installed, since it gives me the same battery life of Catalyst, and I also got that bug with Catalyst (it's a bug in the original AMD code). If you know how can I get logs of suspending / resuming, please, tell me
                      I can confirm Suspend/Resuming not working on my E-350, but it worked with Catalyst.

                      Comment


                      • #26
                        So basically this 3.11 dpm feature was only a usual Phoronix exaggeration?
                        It seems like a half-done alpha code of some sort, nothing complete.

                        (I'm glad it didn't kill my gpu.)

                        Comment


                        • #27
                          Originally posted by Ruse View Post
                          So basically this 3.11 dpm feature was only a usual Phoronix exaggeration?
                          This is the initial release of the code and it's disabled by default as there are bound to be bugs until it gets further testing outside of the cards I used to write it. So it's available to those that want to play with it and when it's ready, it will be enabled by default.

                          Comment


                          • #28
                            Originally posted by Ruse View Post
                            So basically this 3.11 dpm feature was only a usual Phoronix exaggeration?
                            It seems like a half-done alpha code of some sort, nothing complete.

                            (I'm glad it didn't kill my gpu.)
                            You...are a really pathetic troll. Of course the code was going to have bugs in it, the feature is also disabled by default in 3.11 to make sure that end users dont become testers. The only ones who are going to be using this code are the ones who understand the fact that it not considered stable and may not work properly for all GPU's.

                            Comment


                            • #29
                              Originally posted by agd5f View Post
                              This is the initial release of the code and it's disabled by default as there are bound to be bugs until it gets further testing outside of the cards I used to write it. So it's available to those that want to play with it and when it's ready, it will be enabled by default.
                              I didn't blame you or anything like that.
                              Phoronix tends to do exaggerate (like always). "New feature, wowz, year of the linux desktop awesome #omg"

                              It's false advertisement, but not by you.
                              Thanks for your hard work, I'll keep an eye on your drm repo and the mainline kernel as well. (Hope I will be able to help.)

                              @Eric: See my reply.

                              Comment


                              • #30
                                Originally posted by Ruse View Post
                                I didn't blame you or anything like that.
                                Phoronix tends to do exaggerate (like always). "New feature, wowz, year of the linux desktop awesome #omg"

                                It's false advertisement, but not by you.
                                Thanks for your hard work, I'll keep an eye on your drm repo and the mainline kernel as well. (Hope I will be able to help.)

                                @Eric: See my reply.
                                I don't really see how even Michael exaggerated it. The feature is in mainline, that means its available for easy testing. It IS an available feature, its an incredible feature. It just doesnt have all the kinks worked out yet.

                                Comment

                                Working...
                                X