Announcement

Collapse
No announcement yet.

RadeonSI/Gallium3D Still Appears To Have Greater CPU Overhead Than The NVIDIA Driver

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

  • #21
    Originally posted by dungeon View Post

    Not that long time ago in a galaxy far, far away... look at a TF 2 results it is nVidia diff

    In whole, if you compare AMD Linux vs Windows on TF2, you would probably realise it runs at 66.6% of Windows performance, so to not beat but just to equal to Windows we need another 33.33% more perf from somewhere
    That will be a lot more interesting if Valve ever switches windows and linux both to use the Vulkan backend. Until then, you're comparing apples to oranges.

    Comment


    • #22
      Originally posted by Master5000
      Feels good to be me....
      just say no to drugs

      Comment


      • #23
        Originally posted by smitty3268 View Post

        That will be a lot more interesting if Valve ever switches windows and linux both to use the Vulkan backend. Until then, you're comparing apples to oranges.
        Nope, my point was real with real numbers... and it does not depends of ever problematic if statement

        But OK, if that happen we will see - maybe it would change from 60%ish to something like even 80% of Windows on average.

        On the other side if that does not happen, then what? am i right?
        Last edited by dungeon; 23 May 2017, 01:11 AM.

        Comment


        • #24
          Originally posted by dungeon View Post

          Nope, my point was real with real numbers... and it does not depends of ever problematic if statement
          Nope, my point was exactly what i stated, and it does depend on the if statement.

          I don't really care about your "points."

          Comment


          • #25
            Originally posted by smitty3268 View Post
            I don't really care about your "points."
            Cool So why you even comment then

            Comment


            • #26
              Originally posted by dungeon View Post

              Cool So why you even comment then
              To make my own points.

              A better question would be why do I read other people's comments, and that's one I'm wondering myself right about now.

              Comment


              • #27
                Originally posted by phoronix View Post
                Phoronix: RadeonSI/Gallium3D Still Appears To Have Greater CPU Overhead Than The NVIDIA Driver

                In CPU-bound Linux games, the NVIDIA Linux driver still appears to perform better than the newest RadeonSI Gallium3D code...

                http://www.phoronix.com/scan.php?pag...1080p-Overhead
                Driver overhead cannot me measured this way.
                If you want to mesure CPU charge of something, you have to make the something run the same way, and then monitor CPU charge.
                So this eman you should bench using fixed and stable fps. For example, fixing the max fps at 100 in your example. And then monitor general charge individualy and real time on all core/thread, and CPU time used by the driver process, if possible.

                Your graph just show that the NVidia card is faster when GPU bound, but that doesn't mean there is more driver overhead. Juste the card is faster when being CPU limited.

                I don't know the Linux driver at all, but actually, it's most likely there is actually less driver overhead on the AMD platform, because AMD GPU use an hardware scheduler, when NVidia is a full software one.

                Comment


                • #28
                  Originally posted by smitty3268 View Post

                  To make my own points.

                  A better question would be why do I read other people's comments, and that's one I'm wondering myself right about now.
                  In order to have a discussion because this is a forum?

                  Anyway - what I see on all the benches thus far is:
                  - AMD OS driver matches the Windows performance on certain games, esp. if they are native games and not ports
                  - Same with AMD OS vs. Nvidia CS
                  - There are only few native games, most of them are ports (which is why we would still see a performance difference, even if the drivers had 1:1 Windows performance
                  - While still some bugs and performance issues are out in the wild, the out-of-the-box performance for most games is already good and playable
                  - The driver improvements are massive in the last couple of months

                  Personal flair:
                  - The relevance of those benchmarks is heavily biased to ones own lib of games (duh!)
                  - I was able to already convince some folks to try out Linux for gaming and so far, they are happy. They still expect a tad bit more of performance, but are surprised on how well it worked out

                  Comment


                  • #29
                    Originally posted by smitty3268 View Post
                    To make my own points.
                    So what you see wrong on that Arstechica article? They picked mediocre nvidia GPU (also not newest one but couple years existing one with proven stable drivers), mediocre CPU and do benches, for mediocre userbase so called average Joe

                    That is reality and best average showcase of a platform Not "if" statements and not what someone experimentally might get in his "patch me" laboratory and particulary not on the top of the line hardware As these kind of people might get more bump regardless of a platform, so that is again rule of a thumb.
                    Last edited by dungeon; 23 May 2017, 04:50 AM.

                    Comment


                    • #30
                      An abstracted GPU driver will have more overhead.

                      Comment

                      Working...
                      X