Announcement

Collapse
No announcement yet.

Legacy Radeon Performance On Mesa 9.1 Gallium3D

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

  • Legacy Radeon Performance On Mesa 9.1 Gallium3D

    Phoronix: Legacy Radeon Performance On Mesa 9.1 Gallium3D

    While there have already been a number of Radeon Gallium3D benchmarks from Mesa 9.1 using the common R600 Gallium3D driver that supports the Radeon HD 2000 through Radeon HD 6000 series graphics cards, still in existence is the R300g driver. For those still left using a vintage Radeon 9500 (R300) through Radeon X1000 (R500) graphics cards -- basically any ATI GPU roughly between seven and eleven years old -- there's this legacy open-source graphics driver. R300g doesn't see nearly the amount of development activity that the more modern R600g driver sees, but there's still a fair amount of changes. In this article are benchmarks of Mesa 9.1-rc2 on an ATI Radeon X1800XT (R520) graphics card compared to the past four Mesa/Gallium3D stable releases.

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    How do these drivers compare to Windows? I feel like considering their age and relative simplicity, they ought to be about as good as they can get at this point.

    Comment


    • #3
      I wouldnt make the mistake of discounting r300 class harware as simple. It took a long time to come this far and would not have happened without documentation from AMD.

      Comment


      • #4
        What about the performance w/o MSAA enabled, on those games that regressed? Would it have improved w/o it?

        Comment


        • #5
          The Nexuiz performance is sharply lower on Mesa 9.1-rc2 since multi-sample anti-aliasing (MSAA) is finally working for the Radeon X100 GPUs on Mesa 9.1.
          Turn it off so you get a useful graph. It's like this in /every/ article that uses nexuiz. Comparing performance of non-MSAA vs MSAA is silly.

          Comment


          • #6
            Originally posted by schmidtbag View Post
            How do these drivers compare to Windows? I feel like considering their age and relative simplicity, they ought to be about as good as they can get at this point.
            I think we'll find again when he does his Catalyst comparison that there is still a ways to go. On the other hand, OpenGL support and system stability is a lot better on my r515 now than it ever was under Catalyst or the open source drivers 5-7 years ago.

            Comment


            • #7
              Originally posted by schmidtbag View Post
              How do these drivers compare to Windows? I feel like considering their age and relative simplicity, they ought to be about as good as they can get at this point.
              There is absolutely no comparison...
              The final AMD Catalyst drivers for the r300 hardware sucks..
              1. It can't handle WebGL at all without locking up the whole system and so it got blacklisted by all the web browsers so you can't use the GPU for WebGL.. The only way to get good WebGL on the R300 hardware is to run Linux with the Gallium3D drivers. In Windows XP/Vista/7, you can't get GPU accelerated WebGL on R300 hardware no matter what web browser you use, PERIOD.
              2. Incorrect rendering on a lot of games (hammer fight, Heroes of Newerth at highest graphics settings)
              3. Completely Locks the entire system up in Crayon Physics Deluxe (Humble Bundle) every 10 mins or so.
              4. Cannot set PowerPlay mode into "low-power" if your laptop is connected to power. This means your GPU runs at full clocks and your laptop gets blazing hot after a few hours, even when completely idle, whenever connected to power. Gallium lets you set whatever PowerPlay mode you want to use, whenever you want to use it which for me means I can run my laptop for 10+ hours and it's still cold to the touch. There's no reason to have the GPU clocked up high when writing code or e-mails!

              The one thing that it had going for it was MSAA that was all.. Now Gallium for R300 can do that too so there's just no comparison.

              The 33% performance increase in OpenArena sounds pretty sexy, but I noticed a very large performance drop in OpenArena 0.8.8 moving from Catalyst to R300g.. This performance drop didn't impact earlier versions of OpenArena 0.8.5 and I assumed it had something to do with shaders not optimizing well in R300g because OpenArean 0.8.8 brings with it a lot of shaders and "special effects". This 33% performance "gain" may be nothing more than regaining that lost ground and certainly doesn't mean it's faster than the very old Catalyst driver..

              Originally posted by mattst88 View Post
              Turn it off so you get a useful graph. It's like this in /every/ article that uses nexuiz. Comparing performance of non-MSAA vs MSAA is silly.

              Absolutely agreed.. It's Apples and Oranges.. Also, there is another article where he put a graph up and noted that one of the benchmarks ran with rendering artifacts or graphical glitches on a certain version of mesa... Professional benchmarking websites would consider any render artifacts or rendering glitches, no matter how slight they are, a "fail" and would be a 0 FPS. Brtual, but it's a lot better than giving a number such as "200 FPS" when the GPU might not even be rendering half the data.
              Last edited by Sidicas; 19 February 2013, 10:56 PM.

              Comment


              • #8
                Originally posted by Sidicas View Post
                The 33% performance increase in OpenArena sounds pretty sexy, but I noticed a very large performance drop in OpenArena 0.8.8 moving from Catalyst to R300g.. This performance drop didn't impact earlier versions of OpenArena 0.8.5 and I assumed it had something to do with shaders not optimizing well in R300g because OpenArean 0.8.8 brings with it a lot of shaders and "special effects". This 33% performance "gain" may be nothing more than regaining that lost ground and certainly doesn't mean it's faster than the very old Catalyst driver..
                Well, it is definitely true that OpenArena 0.8.8 is a lot more GPU hungry.

                Comment


                • #9
                  Originally posted by Sidicas View Post
                  Also, there is another article where he put a graph up and noted that one of the benchmarks ran with rendering artifacts or graphical glitches on a certain version of mesa... Professional benchmarking websites would consider any render artifacts or rendering glitches, no matter how slight they are, a "fail" and would be a 0 FPS. Brtual, but it's a lot better than giving a number such as "200 FPS" when the GPU might not even be rendering half the data.
                  All of the graphs used are fully-automated from start to finish for maximum reproducibility and streamlined testing. The rendering issues were noted in the article as OpenArena/ioquake3 doesn't provide any convenient mechanism for recording screenshots in a well defined manner for easy comparison to analyze that in an automated way.
                  Michael Larabel
                  https://www.michaellarabel.com/

                  Comment


                  • #10
                    Originally posted by Michael View Post
                    All of the graphs used are fully-automated from start to finish for maximum reproducibility and streamlined testing. The rendering issues were noted in the article as OpenArena/ioquake3 doesn't provide any convenient mechanism for recording screenshots in a well defined manner for easy comparison to analyze that in an automated way.
                    And that is a major weakness in PTS, unfortunately. I'm not sure what the answer is, but if it's "we can't do anything about that" then people are never going to trust the results.

                    Even something as simple as an optional question at the end of the test about whether it completed successfully, so that you could at least fix the test results you post on Phoronix, would be a big help. I mean, you are noting the problem in your article, so obviously you could get that info into PTS somehow.

                    Comment

                    Working...
                    X