Announcement

Collapse
No announcement yet.

Has AMD Finally Fixed Tearing With Its Linux Driver?

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

  • #31
    Originally posted by mugginz View Post
    One thing in AMD's favour though is they seem to have at lest gotten the hairy moths out of the soup at last.
    (Note to self: Don't read one of these threads while taking a drink... )

    ROFLMAO!

    Comment


    • #32
      Originally posted by mugginz View Post
      This is with 3 x 24" LCDs, each @ their native resolutions of 1920x1200 with no rotation or anything fancy.
      You're kidding, right?

      Let's do the math for just precisely what you're doing there:

      1920 x 1200 x 4 x 8 = 73728000 bytes for the raw framebuffer for one screen.

      Approximately 70 Mb.

      Times three is 210 Mb. That's raw framebuffer.

      Now, double buffering's 420 Mb...

      Add in any 3D mode operations such as 2X FSAA and you're at your 1Gb limit.

      "Nothing fancy" indeed.

      Looks like I'll have to dump this adaptor after all. Seems 1 GB isn't enough.
      You're going to find...issues...with anything with 1Gb doing what you're trying to do there. You're needing 1.5-2Gb of card memory to do it, even with NVidia.

      Comment


      • #33
        Originally posted by allquixotic View Post
        If you would prefer that all 3d graphics drivers remain in-house, unreleased research projects until they are 99.99999% perfect, then please delete all instances of Catalyst, the NVIDIA binary driver, and all open source graphics drivers from your PC right now, and do not install them until, oh, 2025 or so at the earliest. For the open source drivers, I'd say 2050. And even then, there will be the occasional engine built with a flaw that leads to a head gasket breach after 50,000 miles, or the occasional hair or fly in your soup.
        The funny thing is that while not having all features implemented yet the open source drivers just "feel" qualitatively so much better. Catalyst can do more but it can't do it very well.

        Originally posted by allquixotic View Post
        Maybe AMD should call the Catalyst Linux driver a "public beta". Would that make you happy?
        Call it public Alpha and I am happy. No, really.

        Originally posted by allquixotic View Post
        A beta driver is like a "soup" scraped up from random kitchen scraps that were slated for the trash, called "chef's experiment" or something on the menu.
        Nowadays many people seem to think that. But that's not really what a beta is. What Firefox does are betas. Well, except for implementing new Features.

        Originally posted by allquixotic View Post
        By setting your expectations right, you don't get a sour attitude when something's in there that you don't find appealing.
        Do you think AMD can't be a competitor to nvidia? Because that's what my expectations are and I'm disappointed.

        Originally posted by allquixotic View Post
        In fact, this whole issue with you seems to be about expectations.
        No, the issues are for example freezing games, laggy overall experience, alt-tab issues etc.


        Originally posted by allquixotic View Post
        Should we hold AMD to blame for putting out a driver before it's ready? It's hard to say, "Well, you should have done better than you have, in less time, and delivered everything I want yesterday!" in an industry that is barely a decade old, to a team of at least 50 people (in the Catalyst case).
        I would blame them for the latter.

        Originally posted by allquixotic View Post
        1. They both support the same level of the OpenGL API. Damn vague Khronos specifications and interpretation mismatches aside, it seems that the few users of the latest OpenGL specs (ahem, Unigine) can run fairly well on Catalyst. And if you go back to, say, OpenGL 2.1, Catalyst does about as well as it possibly can; it's hard to expect more.
        Yes, the rendering itself is actually good. The problems are not there.

        Originally posted by allquixotic View Post
        2. Catalyst supports cross-API, full-screen tear-free, while NVIDIA doesn't (they only support tear-free within an individual API, e.g. OpenGL or composited desktop or a video, but not sync between all three at once). Looks like Catalyst actually wins this battle, now doesn't it?
        If you look at datasheets, yes. In reality, well... Last time I checked I couldn't spot any tearing with nvidia anywhere. Why would I bother if catalyst "wins" with a feature that wouldn't need to be there?

        Originally posted by allquixotic View Post
        3. Multi-monitor is well-supported in Catalyst these days
        You mean, like a friend of mine who hasn't found another solution than connecting the beamer and then restarting X in order to get it to work?
        Why no randr support? Is it too hard?

        Originally posted by allquixotic View Post
        4. Perhaps there are video decoding issues with Catalyst under Xv and XvBA, whereas NVIDIA's VDPAU is supposed to be awesome. This is a fair point; Catalyst video playback could use some work. To me, this isn't an indispensable feature, because using OpenGL for video keeps the hardware acceleration aspect, while basically guaranteeing that a conforming OpenGL implementation will render a pixel-correct result. Since the performance (i.e., the smoothness of the framerate) is acceptable for me even at 1080p, I am satisfied.
        I am not. I'm using the catalyst right now and with vaapi Videos are laggy. I have a HD 4670 and Athlon II 240. That should be more than enaugh for smooth video.
        And even trying with GL output, lol
        Code:
                   ************************************************
                   **** Your system is too SLOW to play this!  ****
                   ************************************************
        Originally posted by allquixotic View Post
        So on point 2, Catalyst "wins", and on point 4, there exists some ancient video API that Catalyst incorrectly implements while it correctly plays back video with another API instead. Boo hoo. From my perspective, Catalyst is actually better than the NVIDIA binary driver as of the 11.1 release.
        Well, it's the best catalyst release since one year or so. I haven't spotted any new bugs so far.
        As usual the newly introduced feature doesn't work right, but that's ok, because it's not enabled by default.

        I wonder why you didn't mention the points where catalyst not only loses but that makes me want to buy a nvidia card next.

        I mentioned the bugs introduces with nearly every new release. Just look at the forums, plenty of it. Vsync with 30 fps and extremely laggy, black artifacts in firefox/pidgin, etc. pp.
        Would you not agree to someone here who said he doesn't dare to press alt-tab on fullscreen-3d-windows because you really don't know what will happen?
        Will it freeze your X? Will it [defunct] it? Will you get "BUG: scheduling while atomic: X/4405/0x10000002"? Anyway you shouldn't change to an maximized window. It will be behind the fullscreen window AND it will have the focus. You then need to use alt-tab blindly (all indication will be hidden behind the fullscreen window) to set the fullscreen window active. Really funny, try it.

        The anti-tearing thing is strange in 2D. 3D works surprisingly extremely well but 2D really feels laggy. The window title bar sometimes has graphics corruption, but it's not too bad.

        As for video acceleration... I said it is laggy. I'm not making this up it's occassionally omitting up to a second of Frames. I wouldn't want to watch a video with this kind of stuttering.

        And then there is the issue of how the graphics feels. Even in warcraft 3 the difference is notably smoother with nvidia. While with fglrx the fps are good, it just does fell better with nvidia. Oh, and in half life 2 the difference is extremely. Really, try it. Well, if half life 2 finally works again. Since 2 catalyst releases it is broken with wine. Near water fps will drop to < 1 fps if you look into the wrong direction and when going near the surface it completely freezes. Well, with OffscreenRendering?fbo it does. With backbuffer the FPS don't drop and it only sometimes crashes under water. But then, most of the game is drawn horizontally mirrored.
        I really encourage you to try it with nvidia. Same configuration, extremely smooth gameplay without issues.
        It may be that nvidia works closer with wine together but if so, AMD SHOULD DO IT TOO!!

        I don't even start to wonder why amdcccle always segfaults on quitting
        Code:
        amdcccle[6819]: segfault at 20 ip 00007fa567cb27a6 sp 00007fff6513db50 error 4 in libc-2.13.so[7fa567c3f000+158000]
        That said, I had none of these issues with a Nvidia Geforce 880GT.
        None.
        The only thing is that it's graphics memory is broken so it will crash if you keep filling it with 3d applications.
        I would prefer using it over my HD 4670 any time. Not because I somehow like nvidia but because it actually has nearly no issues whereas fglrx has many.

        Comment


        • #34
          I'm not sure but I think this is fglrx' fault also:
          http://pastebin.com/GtqiAr9X

          Click on Link in xchat

          receive X freeze

          (yes, I killed all that processes over ssh and screen was still black and the keyboard was still blocked)

          happens only occasionally.

          Comment


          • #35
            Originally posted by Svartalf View Post
            You're kidding, right?

            Let's do the math for just precisely what you're doing there:

            1920 x 1200 x 4 x 8 = 73728000 bytes for the raw framebuffer for one screen.

            Approximately 70 Mb.

            Times three is 210 Mb. That's raw framebuffer.

            Now, double buffering's 420 Mb...
            It's worse than that, since tearfree is (presumably) using triple-buffering, which means 730MB just for the color buffers. Add the necessary z-buffers (another 730MB) and the out-of-memory warning starts making sense now, don't they?

            No wonder that the 6950/6970 cards come with 2GB memory.

            Comment


            • #36
              Originally posted by BlackStar View Post
              It's worse than that, since tearfree is (presumably) using triple-buffering, which means 730MB just for the color buffers. Add the necessary z-buffers (another 730MB) and the out-of-memory warning starts making sense now, don't they?

              No wonder that the 6950/6970 cards come with 2GB memory.
              Double- and triple-buffering do not replicate the depth buffers (or any other intermediate work areas used in rendering), just the colour buffers.

              Comment


              • #37
                Anyone tried it on an up-to-date Arch Linux system yet?

                How is suspend/resume these days with Catalyst?

                Oh and about the buffers and memory usage, how does mirrored screens compare to extended screens?

                Comment


                • #38
                  I tried it now a little bit.

                  Again with 11.1 it is completely impossible to alt-tab from a fullscreen-size 3D-window (for example wine with virtual desktop) to any other window when compositing is enabled. The screen just has a nasty flicker and then the window is back on top. You can't minimize the 3D-Window, you can't get away from it with changing workspaces. It stays in front of everything from start to end. No way of interacting with your OS in that time other than through hotkeys.
                  Without compositing it still stays on top when changing workspaces or trying to minimize it but at least you can alt-tab to little windows. Little in the sense that they are not the size of the screen, because then the same as with compositing and any window happens.

                  Another feature it reintroduces is the framebuffer-hanging. That was one of the bugs introduced last year I didn't mention. It thankfully vanished in later releases but now it is back. It happens often when exiting a 3D-Application that the last frame will stay on your screen. With alt-tabbing in fullscreen-3d-windows it happens too. Fortunately changing to tty1 and then changing back to X usually makes it go away.

                  So yes, this is a regular fglrx release. Not fixing the bugs like alt-tab but even make it worse and reintroduce old bugs that seemed to have been fixed months ago.

                  Comment


                  • #39
                    I installed Catalyst 11.1 on Calculate 11. And I can enable the option for tear free video playback. Graphic card: Radeon HD 3200, 384 MB, resolution 1680x1050

                    Comment


                    • #40
                      Originally posted by bridgman View Post
                      Double- and triple-buffering do not replicate the depth buffers (or any other intermediate work areas used in rendering), just the colour buffers.
                      Ok, make that 210MB for the z-buffer.

                      Comment


                      • #41
                        Originally posted by Svartalf View Post
                        1920 x 1200 x 4 x 8 = 73728000 bytes for the raw framebuffer for one screen.
                        Maybe i missed something but from where does came from the factor 8 ? 32bits color buffer ie RGBA8888 is taking :
                        1920x1200x4 = 9216000 so lets say 10Mbytes of memory

                        It seems you were giving Mbits number, card memory are given in byte last time i heard

                        Comment


                        • #42
                          Originally posted by BlackStar View Post
                          It's worse than that, since tearfree is (presumably) using triple-buffering, which means 730MB just for the color buffers. Add the necessary z-buffers (another 730MB) and the out-of-memory warning starts making sense now, don't they?

                          No wonder that the 6950/6970 cards come with 2GB memory.
                          Heh... I knew I'd understated something somewhere... >:-D

                          Simply put...if you're doing more than one display, you're doing something a bit more aggressive than you think.

                          Comment


                          • #43
                            Originally posted by glisse View Post
                            Maybe i missed something but from where does came from the factor 8 ? 32bits color buffer ie RGBA8888 is taking :
                            1920x1200x4 = 9216000 so lets say 10Mbytes of memory

                            It seems you were giving Mbits number, card memory are given in byte last time i heard
                            Hm... (Note to self: Don't try zinging off a response before going to a meeting...you make dumb mistakes like this.)

                            You're right... But he's still got insufficient memory for attempting a triple-buffered configuration. I just got the calculations off. He's still going to probably need 1.5 Gb at least. Let me mull over the best way to express what I'm trying to say here and I'll put it up here in a bit.

                            Comment


                            • #44
                              Originally posted by BlackStar View Post
                              Yeah, right, you wish. All my nvidia cards tear like hell as soon as I disable compiz/kwin or when I rotate my monitor.
                              Originally posted by grigi View Post
                              Agreed, on my notebook with Nvidia binary drivers, it tears really bad. I also don't run compiz or similar (because it still gets slower over time, so Nvidia hasn't fixed their memory leak issue yet...)
                              The only cool thing of Nvidia is how well vdpau is supported... Then again, VAAPI support is becoming better by the day.
                              It's always hilarious to hear people tell you how nvidia doesn't have issues - until someone who actually uses nvidia under Linux shows up.

                              Comment


                              • #45
                                Originally posted by Joe Sixpack View Post
                                It's always hilarious to hear people tell you how nvidia doesn't have issues - until someone who actually uses nvidia under Linux shows up.
                                Nobody denies that nvidia has issues, they just have less issues the the alternatives and for the most part give a better experience then any of the others.

                                Comment

                                Working...
                                X