Announcement

Collapse
No announcement yet.

NVIDIA's Release Happiness Continues Into April

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

  • #41
    Originally posted by Melcar View Post
    Unigine Tropics
    OpenGL
    1680x1050
    4x AA
    16x AF
    Shaders High
    Textures High
    Filter Trilinear
    Reflection
    Refraction
    Occlusion
    Volumetric

    Ubuntu 8.10
    fps: 25.7
    score: 648

    Windows XP
    fps: 26.6
    score: 670

    Unigine Sanctuary
    OpenGL
    1680x1050
    4x AA
    16x AF
    Shaders High
    Textures High
    Filter Trilinear
    Translucence
    Parallax Mapping
    Occlusion
    Reflection
    Refraction
    Scattering
    Volumetric
    HDR
    DOF

    Ubuntu 8.10
    fps: 28.1
    score: 1190

    Windows XP
    fps: 28.1
    score: 1191

    Now don't you find it odd that similar system (PhenII,[email protected],8800GT first single and then sli) with exception of the GPU get identical results?

    OpenGL
    1680x1050
    4x AA
    16x AF
    Shaders High
    Textures High
    Filter Trilinear
    Reflection
    Refraction
    Occlusion
    Volumetric

    openSUSE 11.1
    fps: 25.7
    score: 648

    Windows XP 64
    fps: 26.6
    score: 669
    Again, you are CPU limited. BTW the same test using the DX renderer in windows yielded an average of 27.1 but with such small variances that is easily attributed to standard deviation due to various background processes, cpu clock drift, etc.

    Comment


    • #42
      [...] My switch to nvidia is triggered by Ubuntu 9.04's release. Nvidia's blob is already compatible with X-Server 1.6, but FGLRX? no f**king way they could accomplish that. OSS? The holy brand new 3D stack still gonna take another year to mature.
      Ubuntu 9.04 comes with an fglrx release that supports XServer 1.6 (R600+ only, IIRC) and open source drivers that support XServer on R100-R700.

      Just pointing this out.

      ATI's OpenGL stack is so broken that it even ignores whether there is a Depth buffer being created or not in glutInitDisplayMode() call.
      I've been using ati hardware since 2003 (R300) and I have never ever seen this - even with their old and ugly OpenGL stack. There are problems when you start pushing the drivers (e.g. try FBO blit with 24/32 bit depth buffers), but ati is not alone in that regard (last time I checked, nvidia cards could not attach a 24bit depth buffer to an FBO, if your context used a 16bit depth buffer and vice versa).

      And again, fglrx is inferior in 2d and video, but performs quite well in 3d. The open-source drivers are better in 2d and video, but lose in 3d. Nvidia drivers tend to be more balanced, but at least with ati you have a choice.

      The grass always looks greener on the other side, but you'll be disappointed if you assume that nvidia drivers are flawless. Check their forums for long-standing issues that have not been fixed.

      Edit: #deanjo:
      score: 669 vs score: 1191

      I wouldn't call that similar.

      Comment


      • #43
        Originally posted by BlackStar View Post
        ...

        Edit: #deanjo:
        score: 669 vs score: 1191

        I wouldn't call that similar.

        That seems to be his Tropics score. Either way, CPU limited or not, the GPU will still work and go under stress. The point I was making was that the driver does perform well in 3D. Besides, if I go by Deanjo's logic everything 3D in Linux is CPU limited and nothing can indicate true GPU performance, which leads me to question why the hell everyone is bitching about poor 3D performance with fglrx (since you know, nothing can show true 3D performance).

        Comment


        • #44
          Originally posted by Melcar View Post
          That seems to be his Tropics score. Either way, CPU limited or not, the GPU will still work and go under stress. The point I was making was that the driver does perform well in 3D. Besides, if I go by Deanjo's logic everything 3D in Linux is CPU limited and nothing can indicate true GPU performance, which leads me to question why the hell everyone is bitching about poor 3D performance with fglrx (since you know, nothing can show true 3D performance).
          Not everything is GPU limited, Games such as ET:QW, Quake4 and even old Doom3 are not CPU limited when ran at higher resolutions.

          Comment


          • #45
            That seems to be his Tropics score.
            Which indicates a large difference, no?

            Not everything is GPU limited, Games such as ET:QW, Quake4 and even old Doom3 are not CPU limited when ran at higher resolutions.
            Parse error. Care to rephrase? (These sentences seem to make sense on their own, but not when put together like this.)

            You say that Doom3 is GPU limited on high resolutions, while Unigine is CPU limited? I really doubt that - Doom3 is primarily CPU bound (shadow volume extrusion on the CPU). Its shaders are simple, the polygon count is low, it really doesn't stress a modern GPU at all. Unigine tropics is *much* harder on the GPU.

            Comment


            • #46
              Originally posted by BlackStar View Post
              Which indicates a large difference, no?
              Not at all look closely at Melcars results, two different benchmarks were ran. My results are from the tropics bench which he got

              Ubuntu 8.10
              fps: 25.7
              score: 648

              Windows XP
              fps: 26.6
              score: 670
              Mine is


              openSUSE 11.1
              fps: 25.7
              score: 648

              Windows XP 64
              fps: 26.6
              score: 669
              You say that Doom3 is GPU limited on high resolutions, while Unigine is CPU limited? I really doubt that - Doom3 is primarily CPU bound (shadow volume extrusion on the CPU). Its shaders are simple, the polygon count is low, it really doesn't stress a modern GPU at all. Unigine tropics is *much* harder on the GPU.
              If doom3 was CPU limited it would show 0 increase in a sli configuration. This however is not true and gains are still found when running a sli rig at higher resolutions (1920x1200 +) with full eyecandy maxed out.
              Last edited by deanjo; 08 April 2009, 02:12 PM.

              Comment


              • #47
                I really like catalyst 9.3 :-) Fast 3d performance (better than my nvidia card). The only thing left is xv under compiz. My 2D seem smooth under compiz, and I can even run every single wine game (EVE, insurgency, cs:source). I really don't see why everyone is bashing fglrx. Yeah true, they have had problems. But almost all the problems have been adressed (for me at least). Fglrx's codebase is quite new, so no wonder why nvidia's blob contains fewer bugs :P

                Anyway. Can't wait for the oss drivers to mature.

                Comment


                • #48
                  Doom 3
                  Ultra
                  4x AA
                  1680x1050


                  Windows
                  122.4 fps

                  Ubuntu
                  127.3 fps


                  Last edited by Melcar; 08 April 2009, 04:07 PM.

                  Comment


                  • #49
                    Originally posted by Melcar View Post
                    Doom 3
                    Ultra
                    4x AA
                    1680x1050


                    Windows
                    122.4 fps

                    Ubuntu
                    127.3 fps


                    You mind posting your method of running the benchmark so I can run the comparison @ home on my system that is setup with nv cards? I'll run it @1680 and 1920.

                    Comment


                    • #50
                      Doom3 is definitely no valid test to compare as you have to force AA/AF within the driver to get used. Inside win the game settings are correctly applied. Take a screenshot to compare.

                      Comment

                      Working...
                      X