Announcement

Collapse
No announcement yet.

Catalyst 10.2 is out

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

  • Yeah, the backclear patch fixes the slowdown without introducing the old-content problem. There have been a few reports about memory leaks but they seem to have had other causes. There was also a report about something in the KDE suite actually relying on the backfill and so not drawing properly with no-backfill or backclear, but I didn't hear anything more about it.

    Once we get the new 2D acceleration implemented in fglrx the patch shouldn't be required, although I hate calling that a solution -- it will just mean that there will be a large and normally wasted copy happening all the time, but all the drivers will accelerate it enough that everyone will live with it.

    Comment


    • Originally posted by bridgman View Post
      Once we get the new 2D acceleration implemented in fglrx the patch shouldn't be required, although I hate calling that a solution -- it will just mean that there will be a large and normally wasted copy happening all the time, but all the drivers will accelerate it enough that everyone will live with it.
      I didn't know in wich reality you live, but in my reality this behavior makes gnome unusable to me. If you read the follow ups to bug 351186 in launchpad, there a many people thinking about it just like me. It's my decision to create the ppa for lucid again. In some other words: With other words - Without this patch linux is worthless for me. If i want use linux in a fast and reliable way without, i have to buy other hardware then ati. i have done so one time and it comes out, that i can live with it for my test-server.

      Nothing to say, i'm not glad about the situation. My cheap asrock board with a G4x on works like hell with compiz. It takes about 40 to buy it. My Gigabyte MA790FXT-UD5P + AMD 965 BE + ATI 4890 takes about 600 and work like shit. (..accelerate it enough that everyone will live with it.) You must be joking.

      'til a half year ago i only use windows and the hardware was great. Then i try linux for some working reasons. I was upset about the poor performance. Then i found a existing ppa and everything was ok for me. I don't mind about intel problems with the patch. If you drive intel, you don't have to apply it. I agree the patch is not the final solution but it solves my problem.

      The other solution is buying different hardware for my laptops and desktops. Without AMD/ATI in it - this will be a solution too. If you have that in mind, tell me...

      The third solution is to fix these f*****g drivers/servers, so they work for both windows and linux. This is what i prefere.

      (I didn't want to start a flamewar, but this is the real world)

      Comment


      • Sorry, I don't understand your rant. The current situation is that on NVidia and Intel hardware (and ATI hardware with open source drivers) you don't see the lag because the drivers accelerate the download. The fglrx driver still uses XAA, so it does not accelerate the download and you see the lag.

        What I'm saying is that once the new 2D accel arrives in fglrx that driver will operate the same way as everyone else w.r.t. this specific issue. Isn't that what you want ?

        Comment


        • $%@#!$#@&$^ edit window...

          The backclear patch allows drivers which don't accelerate that download to run without the lag. Once the new 2D accel arrives, you won't need the patch any more.

          What did you think I was saying ?

          Comment


          • so I hope that your new Direct2D / 2D acceleration approach will accelerate X in most (if not all needed) situations where it's noticable in everyday work

            it's pretty sad to see such bad (slow) gtk, XRender, 2D-acceleration in general with fglrx


            also please try to convince the developers to support newer X-Server and kernel releases more often - 10.4_pre's outstanding support (Lucid Lynx's) surely is a premiere: working with xorg-Server 1.7* and 2.6.34-rc* kernels (only with patches but hey ! it's a start and even better than nvidia )

            also: why is 10.2, 10.3, 10.4_pre causing segfaults on so many cards making acceleration and/or compositing unusable with them for most people ? (error 7 in libGl.so [I don't know it's exact name]) <-- fixing that should be TOP 1 PRIORITY

            thanks !


            P.S.: I love my Cypress

            Comment


            • Originally posted by bridgman View Post
              What I'm saying is that once the new 2D accel arrives in fglrx that driver will operate the same way as everyone else w.r.t. this specific issue. Isn't that what you want ?
              Give open source drivers a try with compiz. You'll see what i mean compared to proprietary fgrlx

              I need my daily eye-candy, wobbeling windows, a 3d-cube... The cheapest intel or nvidia onboard chipsets give that to me. I want to have composite without errors. Is that to hard to understand?

              As i wrote, i'd bought the hardware half year before i start using linux. The system was planned to use with W7 x64 mostly for database conversions and exessive use of virtual machines. It should be able to run a few games on it too (DOOM 3, X3). The nVidia- and ATI-Cards work in a similar performance for a single gpu but the most ati based cards runs cooler and not so noisy. So i decided buying ATI. With the knowledge i have today i'm not shure doing that a second time. Sorry. Under W7 4890 is great, on a unpatched Linux shit.

              The future will bring solutions but i live today.

              Comment


              • I don't think we are communicating

                I understand that there is a performance difference between the open source drivers with EXA and fglrx when running on an unpatched server -- but if you try the open source drivers with XAA acceleration (which is what fglrx implements) then you will find the delay appears there as well.

                Not sure, but I think you are under the impression that I'm telling you that fglrx gives decent performance without the backclear patch, but that is *not* what I am saying. Felix implemented and released the backclear patch back in Sep 2009 because it *was* needed after the previous (no-backfill) patch was *removed* from distributions (yes, removed).

                The new 2D accel architecture will eliminate the need for the patch.

                Comment


                • Originally posted by bridgman View Post
                  The new 2D accel architecture will eliminate the need for the patch.
                  yes because 9-4 beta catalyst do not work at (K/U)buntu10.04 at all.

                  you didn't need to patch the xserver because if you patch it fglrx fails to

                  Comment


                  • The new 2D acceleration is not enabled by default yet; the 10.4 beta is still running the XAA code paths.

                    People are using the driver on Lucid; check the "unminimize" bug report that was mentioned earlier. Some people are seeing a segfault (you are as well, aren't you ?); hoping we can accumulate enough reports to find a pattern and reproduce it in house.

                    Comment


                    • Originally posted by bridgman View Post
                      $%@#!$#@&$^ edit window...

                      The backclear patch allows drivers which don't accelerate that download to run without the lag. Once the new 2D accel arrives, you won't need the patch any more.

                      What did you think I was saying ?
                      Sorry, misread the underlined. My fault. But when will they arrive? I think 'til then the patch is ok.

                      Comment


                      • Originally posted by bridgman View Post
                        The new 2D acceleration is not enabled by default yet; the 10.4 beta is still running the XAA code paths.

                        People are using the driver on Lucid; check the "unminimize" bug report that was mentioned earlier. Some people are seeing a segfault (you are as well, aren't you ?); hoping we can accumulate enough reports to find a pattern and reproduce it in house.

                        I have the segfaulting too (Sony Vaio E, Core i3 330M, 4GB RAM, ATI 5650 1GB) running on Ubuntu 10.4 Beta

                        Comment


                        • Originally posted by TheWretched View Post
                          I have the segfaulting too (Sony Vaio E, Core i3 330M, 4GB RAM, ATI 5650 1GB) running on Ubuntu 10.4 Beta
                          ++

                          ok, let's see what we've got:

                          *) Catalyst 10.2: 2.6.33-zen*, Gentoo ~amd64, gcc 4.4.3 hardened (fstack-protector-all, fPIE), Sapphire 5770, P5W DH Deluxe, Core2Duo E6600; it doesn't matter whether you're using a empty xorg.conf without anything (or little) set at all or full blown with aiglx, texturedvideo, texturedxrender, etc. enabled ==> it's still segfaulting with compiz & kwin (opengl compositing)
                          [<-- that was my old box]


                          *) Catalyst 10.3 [xorg-server 1.6.5] & 10.4_pre [xorg-server 1.7.5] from Lucid Lynx: 2.6.33-zen*, Gentoo ~amd64, gcc 4.4.3 hardened (fstack-protector-all, fPIE), Manufactured by AMD 5850, H57 (Foxconn Bulldog II), Core i7 860
                          [<-- new box]

                          *) also happens on ~x86 (with gcc 4.3.4) on my new box

                          *) I first suspected that it's triggered due to stricter scheduling (BFS CPU scheduler) but it also happens with CFS scheduler


                          summary:
                          happens on
                          *) ~x86, ~amd64
                          *) core2 platform, core i7 platform
                          *) 5770, 5850
                          *) hardened, non-hardened [but both with fPIE at least on my OSes]
                          *) xorg-server 1.6.5, 1.7.5
                          *) Catalyst 10.2, 10.3, 10.4_pre (8.721), 10.3 (with openGL 4.0 support; should be 8.712.3.1)

                          *) according to other reports on Ubuntu 10.04 & various configurations

                          not sure if it can be isolated to 5xxx cards


                          hope that helps !

                          Comment


                          • I just wanted to make it a little bit clearer:

                            segfault in my case means that X works & starts up but apps, that are dependent on opengl / libGl.so.... acceleration (don't know it's exact name) get those segfaults

                            so X == works
                            acceleration & apps needing compositing == don't work

                            examples: compiz, kwin (with opengl compositing; XRender although slow works), glxgears, glxinfo, googleearth, etc. etc.

                            ==> libGL.so.... error 7 segfault or something thereof

                            Comment

                            Working...
                            X