Announcement

Collapse
No announcement yet.

Radeon Gardenshed DRM + Gallium3D Benchmarks

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

  • #16
    Originally posted by bridgman View Post
    - stare at all the things you mentioned
    - get an idea
    - rewrite a bunch of code and see what happens
    - repeat until you have to work on something else

    That said, I believe the main work right now is finishing the enablement of "known" performance-related features, ie the ones which are enabled in r300g but not enabled in r600g. In most cases I think code exists and works on many configurations but not enough to enable by default yet.
    I just wish for the driver to deliver some usable performance. :/
    Its seems I am the only person in universe who wants performance opensource drivers on 2gb 5870 or equivalent amd card. :/
    Even if I switched to gtx260sp216, every time I see awesome amd hardware, I have heart ache. :/
    When I see how slow opensource driver is, I nearly have infarction ://

    Comment


    • #17
      Wow, the HD4670 Gallium3D performance has regressed hard since the last batch of tests! The Catalyst performance on this ASIC has stayed rougly the same. Could this be due to what allquixotic was talking about in comment #16?

      Comment


      • #18
        allquixotic, is there an open bug for the issue you mentioned?

        If not, it shouldn't be too hard to bisect the problem if 7.10 is okay.

        Comment


        • #19
          Originally posted by bongmaster2
          again a useless comparison. my card is way faster tha in the test. vsnc colortiling swap buffers and so on are disabled. it does not matte what is current standard. it does matter whats acually possible with the cards and driver stack. double or triple the r600 bars and u get real results
          I too thought that a hd 4670 was way faster a few weeks before than what I see here.

          So here is what to benchmark next on phoronix: How much speed improvements the mentioned features bring, both individually and combined.
          (color tiling, page flipping, swapbufferwait off etc.).

          Comment


          • #20
            OT - 4770 performance?

            Originally posted by bridgman View Post
            Similar studies have been done before and the conclusion is always the same. Having a higher percentage of the developers dress in black leads to better drivers.

            snip!

            - the shader compiler for r600+ started off as more of a 1:1 IR-to-hardware translator than a real compiler, although recent work may have improved that a lot (haven't had time to look)
            About two years ago I bought a Sapphire 4770, it looked really good on paper (still does) but the stock Fedora sw doesn't impress me particularly, are you able to say what is the state of open drivers for this card?

            Dave

            ps will dress in black if it helps...

            Comment


            • #21
              and the amazing stuff is that my integrated 4250hd card hit 30 fps in ultra on nexuiz at 1440x900x32 lol and my 4850 is always closing the 90 ish fps at 1680x1050x32

              i think there is something nerfing the fps on the phoromatic test cuz the difference is too big

              my 4250 run standard xorg edgers packages with color tiling and swap buffer off

              my 4850 run hand compiled stack with custom cflags including -Ofast (yeap i have gcc 4.6 o_O)

              in both machines i have 2.6.39 kernel hand compiled for preemtive desktop and 1000hz + some goddies that the ubuntu kernel disable by default

              Comment


              • #22
                Originally posted by ChrisXY View Post
                (color tiling, page flipping, swapbufferwait off etc.).
                Is there anywhere documentation about how to enable/disable those things??

                It is so hard to find useful documentation on all the radeon settings/configuration/features.

                Comment


                • #23
                  Something is off.. Quick comparison with my 5750, AMD X2-545, natty with gnome classic (unity disabled), using 2.6.39, with git of mesa/ati/libdrm..

                  http://openbenchmarking.org/result/1...GR-1105271GR49

                  Regards,

                  Comment


                  • #24
                    Originally posted by erpe View Post
                    Is there anywhere documentation about how to enable/disable those things??

                    It is so hard to find useful documentation on all the radeon settings/configuration/features.
                    man radeon

                    Comment


                    • #25
                      Originally posted by whizse View Post
                      man radeon
                      Thanks, too easy.

                      Comment


                      • #26
                        Perhaps someone could benchmark numerous configurations with different non-standard options (Unity on/off, vsync on/off, etc.) to determine how much each helps/hurts and what the real best-case framerate is?

                        Originally posted by RobertCNelson View Post
                        Something is off.. Quick comparison with my 5750, AMD X2-545, natty with gnome classic (unity disabled), using 2.6.39, with git of mesa/ati/libdrm..
                        Emphasis mine.

                        Thanks for posting that -- I just got a 5770 but can't test it yet, and those benchmarks had me a bit worried.
                        Your framerates seem to be increasing with the resolution -- is that usual, or a sign of an error? Your screen resolution is listed as 1280x1024 -- if that's the monitor's limit, would it cause inaccurate results from the higher-resolution tests?

                        If the results are all right, though -- 58% of Catalyst* at 2560x1600 in OpenArena!

                        *on the Phoronix system. I'm assuming Catalyst is doing some AA and anisotropy too.

                        Comment


                        • #27
                          Originally posted by utrrrongeeb View Post
                          If the results are all right, though -- 58% of Catalyst* at 2560x1600 in OpenArena!

                          *on the Phoronix system. I'm assuming Catalyst is doing some AA and anisotropy too.
                          No. OA doesn't even have an option for AA IIRC, and AF is not the default.

                          Comment


                          • #28
                            This comparison seems about right to me. I have this laptop that has both Intel Core i5-450M CPU/GPU and Radeon HD 5650 GPU. I have been using the latest git kernel/libdrm/mesa/ddx code ever since Evergreen support was initially released.

                            So here is my sad comparison:

                            With Intel GPU when Radeon has been soft switched off I get:
                            + Usable desktop and video performance
                            + Acceptable 3d performance (for a low end card)
                            + Usually stable (but not always)
                            + Computer is cool, fans hardly spin
                            - HDMI output is not supported (due to laptop HW limitation?)
                            - Some minor glitches

                            Instead with r600g when Intel has been switched off I get:
                            + Usable desktop and video performance
                            + HDMI works
                            - Always horrible 3d performance, worse than Intel (I test glxgears in fullscreen 1920x1200 and use Blender, usually not games)
                            - 3d is often too unstable for work (Kernel or Xorg crashes while using Blender), but I have seen nearly stable setups too. Mesa 7.10.x could be more stable.
                            - Computer runs hot and fans usually spin loudly (This probably happens on Win7 too. I sometimes try tweaking power_method and power_profile but they don't make enough difference)

                            Obviously most of the time I end up using Intel. But often after git pull I test r600g. Unfortunately I haven't seen it beat Intel yet.

                            Comment


                            • #29
                              Originally posted by utrrrongeeb View Post
                              Perhaps someone could benchmark numerous configurations with different non-standard options (Unity on/off, vsync on/off, etc.) to determine how much each helps/hurts and what the real best-case framerate is?
                              Well compiz is atleast on, i just can't deal with the unity desktop..


                              Originally posted by utrrrongeeb View Post
                              Your framerates seem to be increasing with the resolution -- is that usual, or a sign of an error? Your screen resolution is listed as 1280x1024 -- if that's the monitor's limit, would it cause inaccurate results from the higher-resolution tests?

                              If the results are all right, though -- 58% of Catalyst* at 2560x1600 in OpenArena!
                              [/i]
                              Usually when testing, 1280x1024 always seems to be greater fps then 800x600 in openarena, but something changed in mesa this last week..

                              I can't test the 2560x1600, but i tried an older crt monitor, same results:
                              http://openbenchmarking.org/result/1...GR-1105271GR14
                              (gateway/samsung vx900 1600x1200 max..)

                              Comment


                              • #30
                                Originally posted by RobertCNelson
                                I can't test the 2560x1600, but i tried an older crt monitor, same results:
                                So it tests at 2560x1600 anyway, but the monitor only gets the portion it can display?
                                What's the video interface you're using?

                                Very interesting results, though.

                                In your low-res Nexuiz test, both your results are the same (within a margin of error).
                                In your high-res Nexuiz test, the CRT framerate is down 62%. Could that have to do with more of the screen being displayable?

                                But your OpenArena test shows the inverse -- the CRT framerate is faster by a significant 9 f/s or 13% in low-resolution, faster by 28 f/s or 18% at 1080p, and 27 f/s or 18% at 2560x1600. So the puzzle remains. :/

                                The World of Padman and Urban Terror tests are similar -- the CRT got a framerate increase -- but less dramatic.

                                The low-res Warsow test is unchanged, but at high-res the new test is 81 f/s or 164% faster.

                                Did the tests look all right? (i.e. no serious rendering errors, or hashed garbage?) [I'm assuming you walked away after starting the benchmark, as I'm planning to do next week.]

                                What I'm wondering from this is, would the impressive results hold through for a true higher-res display, or are they a result of rendering a subset of either the high resolution or the game view on a lower-resolution monitor?

                                But hey -- higher framerates = great!

                                Comment

                                Working...
                                X