No announcement yet.

The Most Comprehensive AMD Radeon Linux Graphics Comparison

  • Filter
  • Time
  • Show
Clear All
new posts

  • #31
    i would like to see nvidia vs ati tests some time..with prop drivers ofc
    and from hd 4xxx to hd 5xxx and from nvidia from gtx2xx to gtx 5xx
    wine and native gaming.
    that would be something new at least.. not the same 2x perf difference between catalyst and opensource driver tests again.


    • #32
      Would be great of one of the driver devs could comment on the CPU usage numbers. They surprised me the most.


      • #33
        Thank you Michael Larabel

        I haven't read everything, but I looked at every page. I like the pictures of those cards. I think this was a good idea.
        I believe I encountered a few errors .. I believe they were mostly typos. => "...the 256MHz GDDR3 video memory. " on page 8.

        I liked the graphs too.. I was a bit surprised fglrx is still sometimes 10x as fast as the open source drivers are.
        The power comsumption and GPU temperature comparison was a very good idea too.

        Imho this comparison wasn't bad at all. It wasn't perfect, but since he's done that all alone it's a pretty lot of work.

        Thank you.


        • #34
          Originally posted by allquixotic View Post
          It's sad that even I believe that the community effort around r600g (both AMD employees and volunteers) is the best thing going in the whole world for open source graphics right now, and that effort still is doing very little to achieve performance numbers that approach the capabilities of mid to high-end cards. The effort seems almost futile at times: even when we have significant developer manpower, and official corporate backing, and open documentation, a grab-bag of showstopping issues, from hidden/DRMed hardware (UVD), to software patents, to bottlenecks in the driver architecture that no one can quite figure out how to eliminate, continue to drag down the effort and make it all seem for naught
          I want them to succeed, but the goal is a moving target, and it seems for every two steps we take forward, the goal moves three steps away.
          I think that's the main issue -- at the start of the exercise everyone was excited about having out-of-the-box X drivers and enough 3D to run a modern desktop. Those goals have been pretty much reached, although power management has been more of a challenge than we expected at the start.

          Rather than celebrating that success, however, the expectations have gone way up (GL3/GL4, OpenCL, HW video decode, performance and power management comparable to proprietary drivers) while the manpower has not.

          The corresponding proprietary driver teams are on the order of 100x the size of their open source counterparts and *they* struggle to keep up with ever-increasing expectations, so it stands to reason that the open source teams face all the same challenges. Hardware is continually becoming more complex, the rate of change is continuing to grow, requirements for DRM and system security capabilities are continuing to grow, and so the driver developers task is continuing to get harder rather than easier.

          The places where you feel you are seeing lack of progress are generally places where none of the developers are able to spend time because of other, more pressing goals. There are still more areas requiring work than there are developers, so you aren't going to see continuous progress in all areas - the model is going to be more like a spotlight moving from one area to another over time.


          • #35
            I would pay ?5 for a spreadsheet of card vs test so I can decide whether it's worth replacing my HD3870: it has enough performance for my needs but I'd like something newer which could use less power and run fanless.


            • #36
              Originally posted by Emyr View Post
              I would pay ?5 for a spreadsheet of card vs test so I can decide whether it's worth replacing my HD3870: it has enough performance for my needs but I'd like something newer which could use less power and run fanless.

              Keep your fiver.


              • #37
                Originally posted by mtippett View Post
                OMG! However, the table is an SVG construct, so not easily pastable into a spreadsheet so I can filter it and add in the missing rows... I'll see if I can process the SVG, otherwise will probably have to just type it up again.

                Koomey's law + economics = I should be able to replace my ?120 3870 with something cheaper and more efficient, then downsize my PSU.


                • #38
                  Originally posted by karl View Post
                  I did not pass the first page as instructed by the author.
                  I was before a subscriber but not anymore.
                  I use AdBlockPlus and NoScript and I don't intend to disable them.

                  Question to Michael Larabel: tell me if this requirement (of either Phoronix Premium membership or no add blocking) will stay for future articles so that I know if I should continue to visit your site or not.
                  How is this any more of a "requirement" than any other site that started bitching that nobody wanted to see their skyscraper Microsoft ads and get bombarded with Trash applets and HP crap that covers the page?

                  He could easily go subscription only if a few people using ad blockers are really costing him his pants.

                  That is, if he thinks anyone will pay for his churnalism.


                  • #39
                    OK, framerate is bad with below 30 fps in a tunnel :*( but why is it not in the test?
                    Nexuiz with 'Normal' quality and AMD 6970M

                    [email protected] ~ % glxinfo | grep version
                    server glx version string: 1.4
                    client glx version string: 1.4
                    GLX version: 1.4
                    OpenGL version string: 2.1 Mesa 7.12-devel (git-63b5902)
                    OpenGL shading language version string: 1.20

                    //edit: with 'auto' (not 'mid' as before) as power_profile I get ~60 fps
                    Last edited by disi; 09-19-2011, 03:49 PM.


                    • #40
                      how about simply requiring membership for certain parts of an article ?

                      If requiring a membership to for example see the conclusion or whatever parts one want to hide, perhaps more people will subscribe to see the missing parts ?