AMD Radeon On Linux 6.13 + Mesa 25.0-devel vs. NVIDIA R565 Linux Graphics/Gaming Performance

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts
  • phoronix
    Administrator
    • Jan 2007
    • 67332

    AMD Radeon On Linux 6.13 + Mesa 25.0-devel vs. NVIDIA R565 Linux Graphics/Gaming Performance

    Phoronix: AMD Radeon On Linux 6.13 + Mesa 25.0-devel vs. NVIDIA R565 Linux Graphics/Gaming Performance

    Ahead of the upcoming NVIDIA GeForce RTX 50 series graphics "Blackwell" and the AMD Radeon RX 9070 series "RDNA4" later in the quarter, I figured it would be worthwhile having a dedicated article looking at the latest upstream Linux graphics/gaming performance for current generation NVIDIA GeForce RTX 40 and AMD Radeon RX 7000 series graphics cards. On the AMD side was the near-final Linux 6.13 kernel along with Mesa 25.0-devel for the latest RADV Vulkan and RadeonSI OpenGL driver support while on the NVIDIA side was their current 565 driver release branch.

    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
  • microcode
    Senior Member
    • Mar 2013
    • 2367

    #2
    The takeaway for me is that Counterstrike 2 has AMAZING frame time consistency.

    Comment

    • Danny3
      Senior Member
      • Apr 2012
      • 2387

      #3
      Strange to see that GTA V, a more than 10 years old game, still cannot be played maxed out in 4K @ 120 Hz on any of the GPUs.
      Here goes my dream of ever being able to play it at such quality...
      Hopefully NTSync will be merged this time and maybe it will be able to help this game too.
      Too bad that the power consumption measurements were missing for it.

      Interesting to see that Rx 7800 was a few times pretty close to RTX 4070, which the new AMD GPUS will compete with from what I heard.
      I guess the new AMD GPUs will come just with small improvements over their previous generation, if the bar is so low and the current AMD GPUS are somewhat there.
      That's sad too.

      Comment

      • lichtenstein
        Phoronix Member
        • Nov 2018
        • 84

        #4
        Hey Michael, thanks so much for these tests! It would be really nice if at some point you could do a round of AI benchmarks with CUDA vs ROCM. As a 6900XT owner running local inference more and more (the new DeepSeek-R1 at 32b is insane!) , I'd appreciate them very much, especially with the blackwell cards right around the corner.

        I'm quite aware that the Nvidia cards are faster at this stuff, I'd just like to see by how much (so I can justify going to the "dark side" or not).
        Last edited by lichtenstein; 22 January 2025, 04:30 PM.

        Comment

        • lukewarmtarsier
          Junior Member
          • Oct 2023
          • 11

          #5
          Are there any regressions in 6.13 vs 6.12?

          Comment

          • luanv
            Junior Member
            • Jan 2025
            • 2

            #6
            which settings are used for the gtav benchmark? i have a 9800X3D and a 4070 ti super and i get 120+fps with every setting cranked and 4xmsaa at 4k

            Comment

            • sporif
              Junior Member
              • Jan 2025
              • 1

              #7
              Originally posted by Danny3 View Post
              Strange to see that GTA V, a more than 10 years old game, still cannot be played maxed out in 4K @ 120 Hz on any of the GPUs.
              Here goes my dream of ever being able to play it at such quality...
              Hopefully NTSync will be merged this time and maybe it will be able to help this game too.
              Too bad that the power consumption measurements were missing for it.

              Interesting to see that Rx 7800 was a few times pretty close to RTX 4070, which the new AMD GPUS will compete with from what I heard.
              I guess the new AMD GPUs will come just with small improvements over their previous generation, if the bar is so low and the current AMD GPUS are somewhat there.
              That's sad too.
              You can definitely get 4K 120 fps in GTA V, as long as you avoid really harmful settings like DX10.1 and MSAA 8X. My 7900 XTX goes from an average of 34 fps with the Phoronix benchmark settings to 63fps with DX11, then 105 fps with MSAA 4X. And with DX11 you can enable tessellation which has very little performance impact.

              Comment

              • Daktyl198
                Senior Member
                • Jul 2013
                • 1577

                #8
                Hey Michael, since many of these games default to using FSR or DLSS, I was wondering if you manually disable them or keep them enabled at default settings?

                Comment

                • nuetzel
                  Senior Member
                  • May 2016
                  • 754

                  #9
                  Originally posted by lukewarmtarsier View Post
                  Are there any regressions in 6.13 vs 6.12?
                  Michael

                  Hello Michael,

                  we NEED badly Linux kernel continuous testing, back.

                  6.12.10 -> 6.13.0

                  show ~10 % gfx degradation (vkmark, glmark2 and some what with my test apps on Steam)
                  with my AMD Ryzen 5 3600 (Zen2) / gfx8 (Polaris 20, 8 GB) devel system.

                  Ups, 6.12.8 was all-time high.

                  6.12.9 wasn't available on openSUSE TW,

                  Comment

                  • middy
                    Senior Member
                    • Sep 2013
                    • 232

                    #10
                    With your geo mean of all tests, it would be nice if you split between ray tracing and non raytracing. And maybe also split the synthetic benchmarks too. That's typically how most hardware reviewers do it these days. Like hardware unbox and techpowerup where their geo means are just the actual game results themselves. The gaming benchmarks themselves are more important since they are actual real world examples.

                    But ultimately you do you. Still great results and thank you for doing these tests.

                    Originally posted by nuetzel View Post

                    Michael

                    Hello Michael,

                    we NEED badly Linux kernel continuous testing, back.

                    6.12.10 -> 6.13.0

                    show ~10 % gfx degradation (vkmark, glmark2 and some what with my test apps on Steam)
                    with my AMD Ryzen 5 3600 (Zen2) / gfx8 (Polaris 20, 8 GB) devel system.

                    Ups, 6.12.8 was all-time high.

                    6.12.9 wasn't available on openSUSE TW,
                    I actually was using the 6.13 RC releases with my 7900 XTX because it has the "disable zero rpm" feature for it and I didn't notice any regression at all. I actually did compare too myself running superposition and cyberpunk 2077 and both were identical. Maybe its just for your polaris card.
                    Last edited by middy; 23 January 2025, 01:18 AM.

                    Comment

                    Working...
                    X