Announcement

Collapse
No announcement yet.

Ubuntu 12.10: Open-Source Radeon vs. AMD Catalyst Performance

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

  • Ubuntu 12.10: Open-Source Radeon vs. AMD Catalyst Performance

    Phoronix: Ubuntu 12.10: Open-Source Radeon vs. AMD Catalyst Performance

    With Ubuntu trying to improve their OpenGL driver support state to push the Linux OS as a platform for gaming, Valve going to be promoting the closed-source NVIDIA and AMD drivers on Linux, and various other challenges still turning up for those trying to use the different Linux OpenGL drivers, here are some new benchmarks comparing the open-source Radeon Gallium3D driver against the closed-source AMD Catalyst driver.

    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
    2D

    A few 2D benchmarks would've been nice, to see how much the Catalyst driver sucks for desktop workload.
    Its really sad 2D is ignored in Catalyst for years now, the mode where most users tend to spend their time.

    Comment


    • #3
      Originally posted by Linuxhippy View Post
      A few 2D benchmarks would've been nice, to see how much the Catalyst driver sucks for desktop workload.
      Its really sad 2D is ignored in Catalyst for years now, the mode where most users tend to spend their time.
      Well, they did implement "A2AA" (AMD 2D Acceleration Architecture), previously known as Direct2D for Catalyst. That was supposed to improve 2D performance across the board. It kind of did, but 2D still stinks like rotten cheese on all the proprietary drivers.

      What really galls me about the open source AMD drivers is how much they've stalled on providing proper 3D support for Southern Islands. I thought it was here many months ago, but I was wrong.... it was just a "first cut". AMD doesn't seem to know how to really finish off the support on their open drivers -- they get it to the point where it can render a few triangles, and move on to the next chipset. And I guess the rest of the "community" built up around the open drivers hasn't invested the time this cycle to get SI running.

      Months ago, I was keeping track of the number of days we went before we could get a true and proper composited 3D desktop experience with working (mostly bug-free and performant) 2D and 3D drivers and OpenGL 2.1. I recall that the first mega-patch for SI hit well in advance of how long it took for Evergreen. Well, now it looks like SI will take longer than Evergreen. Definitely not a good sign, guys.

      Anyway, I don't care anymore. I sold my Radeon on eBay and am exclusively using Intel's excellent Ivy Bridge drivers for typical desktop computing with various compositors (Unity, Mutter, Muffin, Kwin). Works like a charm with 2D apps and light OpenGL. I'm using Bumblebee with an Optimus GPU for workloads where I need more oomph. So let's sum up: my typical 2D and composited desktop workloads are silken smooth and run on my processor's tiny graphics cores. My heavy lifting OpenGL 4.2 / OpenCL / CUDA is done by a 96-core NVS 5400M with the Nvidia proprietary driver. I get the benefit of both the Intel open source 2D and the Nvidia proprietary 3D without the disadvantages of either, and seamless power management. I think Torvalds was giving the finger to the wrong company.

      Comment


      • #4
        Originally posted by BO$$ View Post
        Fucking fuck I have a radeon 3400 and updated to 12.10 and now I have to use the radeon open source since the idiots at amd won't provide updates for my 3400.
        Instead of bashing the open-source radeon driver developers, why not bash Catalyst for having phased out support for your Radeon 3400?

        Why the fuck is it so slow? Are really the amd programmers 3 times better than the open source ones? It's an insult. It's three times faster !! What the hell is this? What are they doing to make them so slow? How can they be so incompetent?
        Because AMD has a heap of fulltime developers working on their OpenGL stack (shared between Linux/OSX/Windows), whereas Radeon is mostly developed by spare time hobby

        Comment


        • #5
          Originally posted by Linuxhippy View Post
          Because AMD has a heap of fulltime developers working on their OpenGL stack (shared between Linux/OSX/Windows), whereas Radeon is mostly developed by spare time hobby
          There are 4-5 people working (payed by AMD) on HW support, LLVM and a few more things here and there. Then there are people like marek (one man army of code) and vadim(???) that are just hobbyists. Plus a few more.

          They need more people in mesa to get the drivers to perform better.

          Comment


          • #6
            What is the fundamental reason for the difference?

            Does anybody know what the fundamental reason for the speed difference is? The proprietary driver is at least 3X faster on all tests - that is not something that you can achieve only with incremental code tweaking.

            There must be something fundamental that causes the difference. Clock programming, memory setup, etc?

            Comment


            • #7
              I guess the open driver on Ubuntu has neither PCIe 2.0 turned on nor 2D tiling.

              I'm not sure what's needed to get PCIe 2.0. Maybe kernel 3.6? For 2D tiling, you need xf86-video-ati from git, so the benchmark results aren't so surprising.

              That said, I've got no idea how much the results would be different with everything from git.
              Last edited by marek; 30 October 2012, 06:55 PM.

              Comment


              • #8
                Maybe it has to do with non optimized compiling for the driver and non optimized shader compiler(I think such things have been mentioned before), among other things.

                Comment


                • #9
                  Originally posted by tmikov View Post
                  Does anybody know what the fundamental reason for the speed difference is? The proprietary driver is at least 3X faster on all tests - that is not something that you can achieve only with incremental code tweaking.

                  There must be something fundamental that causes the difference. Clock programming, memory setup, etc?
                  I can't tell you that I know, but the difference is usually very large when it gets into hundreds of FPS. Even very small inefficiencies add up (100 times), so OSS drivers look very bad.

                  With realistic framerates, the difference is not that extreme. With older hardware (using r300g driver), some apps already have parity with Catalyst, some are even faster, and with r600g results were approaching 60-70 percent. Very much dependent on the workload, though.

                  I think that the general consensus is that once the remaining stuff is fixed (HyperZ, 2d tiling, texture tiling, etc, most of this is ready but not on by default), lots of very small tweaks in many places are needed to get more performance out.

                  For what it's worth, my HD4550 routinely outperforms much faster cards from Michael's tests, so the best way to check is to try it yourself.

                  Comment


                  • #10
                    Sure a hd 4550, i have got such a card as well it was never faster than a hd 5670 and he tests a hd 5770.

                    Comment

                    Working...
                    X