Announcement

Collapse
No announcement yet.

Radeon vs. Nouveau Open-Source Drivers On Mesa Git + Linux 4.9

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

  • Radeon vs. Nouveau Open-Source Drivers On Mesa Git + Linux 4.9

    Phoronix: Radeon vs. Nouveau Open-Source Drivers On Mesa Git + Linux 4.9

    With Mesa Git and Linux 4.9 I have done AMDGPU vs. Radeon, AMD OpenGL vs. Vulkan benchmarks, and other tests while in this article is our first side-by-side comparison of AMD Radeon GPUs on their open-source driver against NVIDIA GPUs on Nouveau when using Linux 4.9 Git and Mesa Git as of this week.

    http://www.phoronix.com/vr.php?view=23653

  • #2
    Michael, thanks for taking my recommendation from last time about going to 0a first. Sounds like it wasn't of *too* much help. The last line of the pstate file reports on the current state of things, and may be a good item to include in these benchmarks for improved transparency. As Karol pointed out, what actually tends to happen is that the memory reclocks, but the shader cores don't. (Which is the inverse what I previously suggested.)

    Either way, if you want to make a benchmark that's more competitive on the nouveau end, give https://github.com/skeggsb/linux/tree/linux-4.9 a go. (Those patches aren't actually slated for 4.9 as far as I know. I think there's also a linux tree in Karol's github account as well with just his patches in case Ben decides to force-push that 4.9 branch.)

    Comment


    • #3
      Not sure if these results are comparable to 4.8 since amdgpu has a big performance regression in the memory manager in kernel 4.9 which is slowing things down. I don't know if it's been fixed in mainline yet.

      Comment


      • #4
        well, I wouldn't use my linux tree for everything reliable except I explicitly say so. skeggsb linux 4.9 tree should contain everything afaik it's based on a 10 days old drm-next tree, which is good enough.

        Comment


        • #5
          Thanks for benchmarks.
          And thanks to nouveau developers for trying.
          Fuck you Nvidia.
          Im' gonna sell my GTX 650 and go to AMD
          And I will tell all the people who come to me for computer related advice not to buy the crap you're selling.
          I'm tired of this bullshit.

          Comment


          • #6
            Michael,

            Please hold off any AMDGPU testing (systems with PAT enabled!) on 4.9-rc1 as Marek said, there's a bad regression right now and a solution is being discussed see here: mm: fix cache mode tracking in vm_insert_mixed() breaks AMDGPU [was: Re: Latest testing with drm-next-4.9-wip and latest LLVM/mesa stack - Regression in PowerPlay/DPM on CIK?]

            Comment


            • #7
              Further see for the RFC patches from airlied these may or may not be the final solution and the Intel folks have a solution also.

              https://lists.freedesktop.org/archiv...ber/002978.htm

              Comment


              • #8
                Originally posted by Danny3 View Post
                Thanks for benchmarks.
                And thanks to nouveau developers for trying.
                Fuck you Nvidia.
                Im' gonna sell my GTX 650 and go to AMD
                And I will tell all the people who come to me for computer related advice not to buy the crap you're selling.
                I'm tired of this bullshit.
                What a coincidence, I have a 650 with nouveau too. And I'm considering the AMD 480 as it will be a big performance jump for me while still having open source video drivers.

                I also thank the nouveau developers, for doing this as it was probably one of the reasons why AMD develops official open source drivers for their Radeon GPUs. Because the incredible effort and courage from this people diving into completely unknown territory, trying to understand all this low level graphic stuff, just to make an open source driver, and that people use it despite the low performance proves without question that people want open source drivers. So thanks to the nouveau developers

                The problem is that now some companies are providing open source """drivers""" that are only some sort of bind between the firmware and the kernel, so the """driver""" doesn't do anything important, the firmware does, which they maintain closed source. They might even cryptographically sign the firmware so no one is able to change the firmware even if they could somehow understand it and change it.

                Comment


                • #9
                  I don't know if it's related to the regression in 4.9 but the cpu overhead of radeonsi looks pretty bad.

                  I mean, it's expected from radeonsi to not have as good a cpu efficiency as the nvidia binary driver yet (with focus rightfully being on GL features).
                  But seeing nouveau as far as 50% ahead (the CS:GO test) is really shocking.

                  I thought mesa drivers shared a lot of code, how is such a big difference possible?

                  Comment


                  • #10
                    radeonsi in these tests is more cpubound than nouveau, i hope it can be explained by kernel regression

                    Comment

                    Working...
                    X