Announcement

Collapse
No announcement yet.

Radeon RX Vega Performance With Mesa 17.3-dev + LLVM 6 + drm-next-4.15-dc

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

  • Radeon RX Vega Performance With Mesa 17.3-dev + LLVM 6 + drm-next-4.15-dc

    Phoronix: Radeon RX Vega Performance With Mesa 17.3-dev + LLVM 6 + drm-next-4.15-dc

    Ending out September is the very exciting news that AMDGPU DC display code will likely land in Linux 4.15. Out of this excitement of finally seeing a mainline Linux kernel with modern Radeon GPUs supporting HDMI/DP audio, atomic mode-setting, and more. I decided to see how well this "drm-next-4.15-dc" code is working out for Radeon RX Vega graphics cards, where attached monitors can finally be driven by this DC code rather than having to rely upon AMDGPU-PRO or other kernel branches. So for ending out this exciting month, here are some fresh benchmarks of the RX Vega 56 / RX Vega 64 and other Radeon GPUs using this kernel paired with Mesa 17.3-dev Git built against LLVM 6.0 SVN compared to various NVIDIA Pascal graphics cards.

    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
    What is happening to Metro: Last Light Redux and Triangle? The performance was much better 15 days ago...

    Comment


    • #3
      Michael It might be interesting to test if primitive binning helps or hurts on Vega.

      Primitive binning can be disabled by adding "R600_DEBUG=nodpbb" into /etc/environment and rebooting.

      Comment


      • #4
        AMD seems to not be able to get there on the gaming market. I was considering buying an rx580 with a ryzen full-amd desktop but just looking at this comparison and the prices of geforce 1060 vs rx 580 makes it a very bad choice. Paying 10-15% premium for a 10-20% lower performance + almost-OSS drivers seems like an overkill, even for a Linux/OSS enthusiast.

        Good to see how the vega will perform in 2018 too...

        Comment


        • #5
          Lukewarm performance really. RadeonSI needs a bit more work. Performing well in a couple of titles at least.

          Comment


          • #6
            Originally posted by Almindor View Post
            AMD seems to not be able to get there on the gaming market. I was considering buying an rx580 with a ryzen full-amd desktop but just looking at this comparison and the prices of geforce 1060 vs rx 580 makes it a very bad choice. Paying 10-15% premium for a 10-20% lower performance + almost-OSS drivers seems like an overkill, even for a Linux/OSS enthusiast.

            Good to see how the vega will perform in 2018 too...
            What do you mean by "almost-OSS"? Talking about the card firmware?

            Comment


            • #7
              Still room for improvement, but I know how far the drivers have progressed these last few years so I can really see the gap closing. Also with DC hopefully being merged into the kernel very soon. It gives me hope we will see AMDs Vulkan driver finally opensourced and features like Freesync, and maybe a proper radeon settings like windows users enjoy (I can dream, can't I? :-p)

              Comment


              • #8
                Originally posted by Almindor View Post
                Aalmost-OSS drivers seems like an overkill, even for a Linux/OSS enthusiast
                you seem to have strange definitions of "oss" and "enthusiast"

                Comment


                • #9
                  What each and every one of you are missing is that, for some reason, a i9-7980XE was used in these benchmarks instead of a better "gaming" (fuck, I hate this word) CPU, like the 7700k. That could have changed some of the results.

                  Comment


                  • #10


                    Try undervolting Vega 56 like this Vega 64 doesn't worth it. You people should also know that AMD obfuscates that the ASIC quality difference exist between the standard and the lasercut part where the smallest one has more quality because you reduced the bad parts. It doesn't exist to a good extend between the lasercut downed versions.

                    Comment

                    Working...
                    X