Announcement

Collapse
No announcement yet.

This Week's Mesa 17.1-dev + Linux 4.11 Radeon Performance vs. NVIDIA

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

  • This Week's Mesa 17.1-dev + Linux 4.11 Radeon Performance vs. NVIDIA

    Phoronix: This Week's Mesa 17.1-dev + Linux 4.11 Radeon Performance vs. NVIDIA

    Given all the recent performance work that's landed recently in Mesa Git for Mesa 17.1 plus the Linux 4.11 kernel continuing to mature, in this article are some fresh benchmarks of a few Radeon GPUs with Mesa 17.1-dev + Linux 4.11 as of this week compared to some GeForce graphics cards with the latest NVIDIA proprietary driver.

    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
    I really wish they'd release the latest version of Tomb Raider, seems that's the best game for AMD users under Linux atm. Nothing really exciting happening on the AMD front until next month with the new Polaris 500 series which are only mild upgrades from the previous.

    Comment


    • #3
      Nvidia still has a huge advantage on csgo at 1920x1080, this should mean mesa drivers have a big cpu overhead when compared to the closed ones.
      Thanks for providing these lower resolution benches instead of just 4K.

      Comment


      • #4
        Originally posted by Max Payne View Post
        Nvidia still has a huge advantage on csgo at 1920x1080, this should mean mesa drivers have a big cpu overhead when compared to the closed ones.
        Exactly. Not just CS:GO. That's perhaps the biggest issue with RadeonSI. When the cards are not limited by the CPU RadeonSI seems to do really well at it's current state.
        But the CPU limit is especially unfortunate because there is a lot of power in the AMD cards that doesn't get utilized in these situations.

        Comment


        • #5
          Originally posted by Max Payne View Post
          Nvidia still has a huge advantage on csgo at 1920x1080, this should mean mesa drivers have a big cpu overhead when compared to the closed ones.
          Thanks for providing these lower resolution benches instead of just 4K.
          CS:GO runs better in wine-staging +nine. The native version uses toGL library, a crap that causes more overhead.

          Comment


          • #6
            It looks like RX480 is competitive with 1060, more or less. Ok, maybe several fewer FPSs in some games, equal or more in others, and drivers for AMD will get better with time.

            Overall, it looks like a good deal. Price wise pcpart picker lists RX480 for 200 GBP vs 1060 for 215 GBP. And you get open-source drivers.

            Comment


            • #7
              Originally posted by prazola View Post

              CS:GO runs better in wine-staging +nine. The native version uses toGL library, a crap that causes more overhead.
              dude, if this was the only issue then proprietary nvidia driver wouldn't be performing better than Mesa...

              Comment


              • #8
                Originally posted by TheOne View Post

                dude, if this was the only issue then proprietary nvidia driver wouldn't be performing better than Mesa...
                But it's one of the problem if the windows client runs better than the native one...or am I wrong? I don't think any sort of translation layer can boost performances.
                Nvidia goes better because the driver is better at multithreading.

                Comment


                • #9
                  I'm sure this is perhaps the main reason why Nvidia just can't open source their drivers:
                  This image shows how many draw calls a GTX 980 can do with DX12 and DX11 with multi threading disabled and enabled. What we can basically notice: The GTX 980 benefits from multi threading even in DX11 and gets up to 200% draw calls depending on the count of active cores.

                  In opposition this is the result of an R9 290X. Although the main purpose of this comparison was the demonstration how much better Mantle/DX12 is it also shows that the results of the R9 290X stay the same in DX11 no matter how many threads are active.

                  So these images look like a proof that Nvidia has hidden and superior techniques enabled to spread draw calls over more threads in their closed drivers causing a higher performance in situations when AMD cards are already limited by the CPU performance. I expect their OpenGL drivers to be optimized the same way.

                  Of course Vulkan is still the future but the current RADV implementation seems to be even more CPU limited than RadeonSI and OpenGL will of course stay very important for many years. So perhaps the key is to find out how to achieve the same results like Nvidia on Windows on Linux with OpenGL. When it could be achieved and AMD wouldn't have fixed it already on Windows I imagine a lot of Linux vs. Windows performance comparisons on youtube being won by Linux and it would surely also help the Fury cards especially.

                  Comment


                  • #10
                    lol.. the issue is 215fps vs 280 fps... 1st world problem especially on 60hz monitor. Amd open source is almost on par with nvidia closed at performance but has other bigger advantages. So looks like a first win for AMD

                    Comment

                    Working...
                    X