Announcement

Collapse
No announcement yet.

Open-Source Radeon Performance Is Very Good For DiRT Rally On Linux

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

  • #21
    I don't know if this is the same reason that the performance has regressed in this test, but I have found that lately I need to switch the CPU and GPU power profiles to performance to get the same performance that I did in the past on default settings.

    With RX480 and 6700k on mesa git, llvm 3.9, linux 4.10, intel_pstate:
    This shader[1], on stock settings (powersave governor, dpm auto) I get 20fps. Once I switch over to performance[2], then I'm up at 60fps.

    I first noticed this problem in games, but it's easier to see on shadertoy. Does anyone else have to do this as well?

    [1] https://www.shadertoy.com/view/4ttSWf
    [2] $ sudo sh -c "echo 'high' > /sys/class/drm/card0/device/power_dpm_force_performance_level; cpupower frequency-set -g performance"

    Comment


    • #22
      Originally posted by ItzExor View Post
      I don't know if this is the same reason that the performance has regressed in this test, but I have found that lately I need to switch the CPU and GPU power profiles to performance to get the same performance that I did in the past on default settings.
      My bug report is maybe related: https://bugs.freedesktop.org/show_bug.cgi?id=99967

      Comment


      • #23
        Originally posted by haagch View Post
        Mine does scale up, I think the load is just not high enough overall and both governors interact badly with each other or something. For that particular shader I mentioned above, If I set CPU to performance then GPU clocking trends upwards a bit and I go from 20 to around 40FPS. If I leave CPU on powersave and only set GPU to max state, then the CPU trends up a bit as well and I'm at around 40fps again. If I enable both, they both sit at their max state all the time and I get around 60FPS.

        Not every game/load changes that much. Games with higher load don't seem to change very much with these settings. Old games running on wine w/ gallium nine seem to get huge slowdowns if I leave it on standard. In CS:GO my fps varies wildly without this, but still averages pretty high > 200 fps.

        Comment


        • #24
          Originally posted by Nille_kungen View Post
          Will there be an RX 580 or RX VEGA?
          The RX 500 series comes out first, then a month later you could expect the RX VEGA to appear, which from the looks of it is going to have 3 flavors on top of the RX 580 performance level, still a bit confused exactly how their going to name them...

          Comment


          • #25
            Originally posted by theriddick View Post

            The RX 500 series comes out first, then a month later you could expect the RX VEGA to appear, which from the looks of it is going to have 3 flavors on top of the RX 580 performance level, still a bit confused exactly how their going to name them...
            I'm not to sure about that.
            If Vega is to be called RX Vega then i think it would make more sense to call the rebranded cards RX Polaris if there will be any rebranded cards..
            In the youtube link i posted it makes me think that there might not be any rx500 series at all.
            It will be interesting to see how it turns out.

            Comment


            • #26
              Originally posted by theriddick View Post

              The RX 500 series comes out first, then a month later you could expect the RX VEGA to appear, which from the looks of it is going to have 3 flavors on top of the RX 580 performance level, still a bit confused exactly how their going to name them...
              This. AFAIK the 500 series are Polaris+ and the Vega lineup will be called RX Vega much as we have the R9 Fury.

              Comment


              • #27
                What is missing is maybe a Linux vs. Windows comparison here.
                This is especially interesting as there are different minimum hardware requirements between the Linux and the Windows version.

                E.g. while a Radeon 6870 should work fine on Windows, the game would probably not even start on Linux with the r600g driver.
                Also heise.de (link in German) tested a Bristol Ridge APU (A10-9700) to get 35 fps on Windows on 1080p with medium settings. I wonder if the Linux version would perform similarly.

                Comment


                • #28
                  I downloaded llvm 3.9.1 from http://releases.llvm.org/download.html#3.9.1 (careful: the debian binaries use the old c++ ABI, the ubuntu ones the new one) and build mesa the mesa 13.0.4 commit with it.
                  OpenGL renderer string: Gallium 0.4 on AMD POLARIS10 (DRM 3.11.0 / 4.10.0-g8fc7b85a6845, LLVM 3.9.1)

                  But I can't see much of a performance difference to my current setup with latest llvm svn and mesa git, around 80 fps on medium (I only tried actual gameplay, no the benchmark).
                  OpenGL renderer string: Gallium 0.4 on AMD POLARIS10 (DRM 3.11.0 / 4.10.0-g8fc7b85a6845, LLVM 5.0.0)
                  kernel is drm-next-4.12-wip

                  Has anyone tested this on archlinux or so and got the "regression" reversed? I wanted to bisect but if I can't get the good performance, not much luck...

                  edit: Okay, I think with echo high > /sys/class/drm/card0/device/power_dpm_force_performance_level I have some performance difference. It's not that big though, more like 80->100 fps. Still, only ingame, not benchmark. Maybe I should run that.
                  Last edited by haagch; 03 March 2017, 07:50 PM.

                  Comment


                  • #29
                    Here are my results. Note that I have a much slower CPU than Michael. It's Core i5 3570. The GPU is Radeon Fury X.

                    I was running the benchmark at 1920x1080, Ultra settings.

                    Mesa 13.0 + LLVM 3.9: 69.6 FPS
                    Mesa 17.1-git + LLVM 3.9: 69.6 FPS
                    Mesa 17.1-git + LLVM 4.0: 70.4 FPS
                    Mesa 17.1-git + LLVM 5.0svn: 71 FPS

                    Comment


                    • #30
                      Bought it because of linux support

                      But unplayable on my favrourite linux platform
                      Code:
                         PROCESSOR:          AMD Phenom II X6 1055T @ 2.80GHz (6 Cores)
                          Core Count:       6
                          Extensions:       SSE 4a
                          Cache Size:       512 KB
                          Microcode:        0x10000dc
                          Scaling Driver:   acpi-cpufreq schedutil
                      
                        GRAPHICS:           MSI AMD Radeon RX 460 2048MB
                          OpenGL:           4.5 Mesa 17.0.2 Gallium 0.4 (LLVM 3.9.1)
                          Display Driver:   modesetting 1.19.3
                          Monitor:          2963
                          Screen:           2560x1080
                      
                        MOTHERBOARD:        ASUS M4A88TD-V EVO/USB3
                          Memory:           12288MB
                          Chipset:          AMD RS880
                          Network:          Realtek RTL8111/8168/8411
                      
                        DISK:               2 x 500GB Western Digital WD5000AAKS-0 + 64GB KINGSTON SNV425S + 63GB SanDisk SDSSDP06 + 2000GB Expansion Desk
                          File-System:      ext4
                          Mount Options:    data=ordered relatime rw
                          Disk Scheduler:   CFQ
                      
                        OPERATING SYSTEM:   Arch Linux
                          Kernel:           4.10.6-1-ARCH (x86_64)
                          Desktop:          KDE Frameworks 5
                          Compiler:         GCC 6.3.1 20170306
                      So sadly no brainer to play it on win 10 as far as I am concerned ... for now!
                      Last edited by kamelie1706; 03 April 2017, 02:11 AM.

                      Comment

                      Working...
                      X