Announcement

Collapse
No announcement yet.

Marek Threads RadeonSI Gallium3D, Big Performance Gains For Many Games

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

  • #11
    Originally posted by boxie View Post
    I guess we both need to see some empirical evidence then!
    Perf is not is specs, so no evidence needed - it is up to designer

    Comment


    • #12
      AMD battles for the future

      Comment


      • #13
        Could someone please point out the essential differences between this work and the recently landed OpenGL threaded dispatch support? I didn't get it.

        Comment


        • #14
          Originally posted by jf33 View Post
          Could someone please point out the essential differences between this work and the recently landed OpenGL threaded dispatch support? I didn't get it.
          This is gallium-threaded, should speed up non games also something that works for all applications.... to say, even your Desktop should be faster
          Last edited by dungeon; 11 May 2017, 02:23 AM.

          Comment


          • #15
            I am looking forward to the +8% in CS:GO ! Right now my fps is frequently dipping under the refreshrate of my display. these additional 8% might help to solve that Thank you very much Marek!!!

            Comment


            • #16
              Is this the same thing that is on AMDGPU? because I think I'm using amdgpu with my furyx card so not sure if I already benefit from this.

              Comment


              • #17
                Originally posted by theriddick View Post
                Is this the same thing that is on AMDGPU? because I think I'm using amdgpu with my furyx card so not sure if I already benefit from this.
                AMDGPU is the kernel part and radeonSI is the userspace part so yes your driver will use it.

                Comment


                • #18
                  Then I understand that this optimization will affect especially those who have a less powerful cpu (for example my fx6350 with many cores but little performance per core), right?

                  Comment


                  • #19
                    The power usage shouldn't be affected much. Please note that even though more CPU in total is used, it's divided into multiple cores so that that the max frequency needed (and consequently the voltage level) stays lower.

                    As for a difference between this and threaded OpenGL, it is mostly about the layer that is threaded. Doing it at OpenGL api layer is potentially more adventageus since the state tracker work is also offloaded to another thread. However, it is more complex and bug-prone and probably will never be enabled by default. The gallium level split on the other hand can be enabled universally.

                    Comment


                    • #20
                      Good job Marek!

                      Comment

                      Working...
                      X