Announcement

Collapse
No announcement yet.

AMD A10-7870K Godavari: RadeonSI Gallium3D vs. Catalyst Linux Drivers

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

  • AMD A10-7870K Godavari: RadeonSI Gallium3D vs. Catalyst Linux Drivers

    Phoronix: AMD A10-7870K Godavari: RadeonSI Gallium3D vs. Catalyst Linux Drivers

    Last week I started posting AMD A10-7870K Linux benchmarks for this "Godavari" APU that's effectively a Kaveri Refresh and slightly faster for its four CPU cores and Radeon R7 Graphics over the former high-end Kaveri, the A10-7850K. In today's articles are some benchmarks of the Radeon R7 Graphics on the A10-7870K when running Ubuntu and testing the open-source RadeonSI Gallium3D driver against Catalyst on Linux.

    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
    no Metro redux tests?

    Comment


    • #3
      Originally posted by madjr View Post
      no Metro redux tests?
      The open source drivers still can't run them. OpenGL 4 and all that.

      Comment


      • #4
        Interesting... the performance offset between all-open and Catalyst seems more consistent between applications than it has been on dGPU.

        Not sure what that means yet but it seems like useful information
        Test signature

        Comment


        • #5
          Originally posted by xeekei View Post
          The open source drivers still can't run them. OpenGL 4 and all that.
          With Airlie's shader subroutine branch it works: https://www.youtube.com/watch?v=e97hP1ys-7s

          A few shaders don't compile, but in #dri-devel it looks like it should actually work and will be fixed quickly. In any way, it already looks okay-ish.

          Comment


          • #6
            Originally posted by bridgman View Post
            Interesting... the performance offset between all-open and Catalyst seems more consistent between applications than it has been on dGPU.

            Not sure what that means yet but it seems like useful information
            I thought that was weird too. I have a feeling the memory has something to do with it. To my knowledge, that's the only difference in the hardware. I wonder if Catalyst is designed to work around the bandwidth limitations.

            I'm not sure what GPU the 7870k uses, but I'd be interested in seeing mesa vs catalyst results for the discrete counterpart.

            Comment


            • #7
              Oibaf drivers currently use LLVM 3.6.1, i believe.

              I'm not sure what all is in 3.7 that's not in 3.6.1 - were all the performance boosting additions backported?

              Comment


              • #8
                Originally posted by smitty3268 View Post
                Oibaf drivers currently use LLVM 3.6.1, i believe.

                I'm not sure what all is in 3.7 that's not in 3.6.1 - were all the performance boosting additions backported?
                AFAIK, main performance improvment comes with scheduler in 3.7, that is not or probably can't be backported... Expect plus 10-15% more fps in Unigine benchmarks with that, and probably in same other cases a few % too, but of course not in all games.
                Last edited by dungeon; 17 June 2015, 09:12 PM.

                Comment


                • #9
                  Originally posted by schmidtbag View Post
                  I'm not sure what GPU the 7870k uses, but I'd be interested in seeing mesa vs catalyst results for the discrete counterpart.
                  The R7 250X (Cape Verde) is probably the closest fit; Bonaire (7790 / 260X) is the same HW generation as Kaveri but almost 2x the shader count.

                  Originally posted by dungeon View Post
                  AFAIK, main performance improvment comes with scheduler in 3.7, that is not or probably can't be backported... Expect plus 10-15% more fps in Unigine benchmarks with that, and probably in same other cases a few % too, but of course not in all games.
                  That makes sense -- the shader compiler difference would make for a fairly constant performance delta with the tested games.
                  Test signature

                  Comment


                  • #10
                    Originally posted by bridgman View Post
                    That makes sense -- the shader compiler difference would make for a fairly constant performance delta with the tested games.
                    Of course it is good but that is just bit of it, other then regular: missing features and optimisations for particular functions, cpu overhead, bandwidth and BO managment is main a difference between all-open, catalyst and even nvidia if you want

                    Take one game as Xonotic for example: on radeonsi APUs are fine on 'low' settings, actually even faster then flgrx, on medium settings too, on 'normal' it is on par... once we came to more demanding area started with 'high' fglrx became faster, etc...

                    So for Xonotic started from high, but also ultra and ultimate... it just cames down to 'motion blur', 'bloom' and 'occlude', with any of those mesa starts to struggle more then fglrx
                    Last edited by dungeon; 17 June 2015, 10:08 PM.

                    Comment

                    Working...
                    X