Announcement

Collapse
No announcement yet.

Radeon RX 6600/6700/6800 XT: RADV vs. PRO Vulkan Driver Performance

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

  • Radeon RX 6600/6700/6800 XT: RADV vs. PRO Vulkan Driver Performance

    Phoronix: Radeon RX 6600/6700/6800 XT: RADV vs. PRO Vulkan Driver Performance

    With yesterday's launch day Radeon RX 6600 XT Linux review the benchmarks were conducted using the popular Mesa RADV open-source driver used by many Linux gamers considering it's the driver Valve has been relentlessly optimizing and is the default on most (or all) Linux distributions. For those wondering how the performance of RADV is comparing to that of AMD's closed-source "PRO" Vulkan driver distributed as part of the "Radeon Software for Linux" package, here are some benchmarks exploring that difference.

    https://www.phoronix.com/vr.php?view=30434

  • #2
    Congrats to the RADAV developers!

    When taking the geometric mean of all the Vulkan benchmarks that were successfully tested on both drivers, the RX 6600 XT and RX 6700 XT were both around 4% faster on Mesa 21.3-dev's RADV than the Radeon Software for Linux 21.30 PRO packaged driver stack while the higher-end Radeon RX 6800 XT was 7% faster on RADV when tested on Ubuntu 20.04 LTS.
    So when the slow code path gets removed, the PRO driver should have a lead, albeit not a big one. This is really astonishing. Who would have thought the free driver would become that good?

    Comment


    • #3
      What does the pro version even exist for? Just so people can play quake ii rtx demo?

      Isn't it just a duplication of effort? They should just scrap the pro driver and focus on mesa.

      Comment


      • #4
        I'm happy to see PRO and RADV getting closer and closer in features and performance. This means that the OSS and proprietary are functioning in very similar ways, which speaks to the small volume of secrets being protected by the proprietary veil.

        It also warms my heart to see AMD continuing to prove their prior attitude - that OSS wasn't worth supporting because it could never result in a viable, performant driver - was entirely wrong. We're happily eating that humble pie and we should be constantly grateful that they had the humility to re-evaluate their position, whereas nVidia (and others) are still on their high, proud, horses.

        It's also worth mentioning that from Polaris backward, AMD's drivers were a disaster at launch and took in many cases upwards of a year to outperform the previous generation. Vega and RDNA have been an entirely different story, with AMD's drivers being stable and performant from day 1. Whether this has anything to do with their OSS efforts, I don't have the insider angle to speak to, but I like to believe that it has, because working with OSS drivers and all those public eyeballs encourages healthy collaboration. That must have positive effects inward.

        Comment


        • #5
          Originally posted by rabcor View Post
          What does the pro version even exist for? Just so people can play quake ii rtx demo?

          Isn't it just a duplication of effort? They should just scrap the pro driver and focus on mesa.
          Why maintain both? Because the OSS version is an exercise in team building and the proprietary version is an exercise in keeping cards close to the chest.

          AMD can get the best of both worlds by doing both at once, measuring the relative successes, and directing resources toward what's most successful.

          It's naive to believe that one path is the best path. Without taking multiple paths, the only thing served by believing you're on the correct one, is hubris.
          Last edited by linuxgeex; 11 August 2021, 05:20 PM.

          Comment


          • #6
            Originally posted by rabcor View Post
            What does the pro version even exist for? Just so people can play quake ii rtx demo?

            Isn't it just a duplication of effort? They should just scrap the pro driver and focus on mesa.
            Working on vulkan in mesa requires hiring more people. As has been stated before, dropping AMD vulkan on Linux does not suddenly free up a lot of resources that can work on Mesa. AMD vulkan still needs to be supported on other OSes.

            Comment


            • #7
              Originally posted by agd5f View Post

              Working on vulkan in mesa requires hiring more people. As has been stated before, dropping AMD vulkan on Linux does not suddenly free up a lot of resources that can work on Mesa. AMD vulkan still needs to be supported on other OSes.
              And no different than having Zen3 / Zen4 teams working at the same time, having OSS and proprietary teams working at the same time provides opportunities to explore what can be achieved in both environments. ie if AMD invents a technique they don't want nVidia to immediately have access to they can implement it on the proprietary side to get some value from it in the short term. If they have new ideas that require a large depth of testing to tune, they can try it on the OSS side where users won't return their cards because their game stopped working... they'll just roll back to the stable repo, and hopefully provide some quality, free, debug feedback.

              Comment


              • #8
                Originally posted by linuxgeex View Post
                It's also worth mentioning that from Polaris backward, AMD's drivers were a disaster at launch and took in many cases upwards of a year to outperform the previous generation. Vega and RDNA have been an entirely different story, with AMD's drivers being stable and performant from day 1. Whether this has anything to do with their OSS efforts, I don't have the insider angle to speak to, but I like to believe that it has, because working with OSS drivers and all those public eyeballs encourages healthy collaboration. That must have positive effects inward.
                If you check the Phoronix reviews, I think you'll see that Polaris had a successful launch as well - we had started the OSS effort with the R3xx-5xx parts (since each new generation is based on the previous one) and spent some years catching up with the introduction of new chips.

                https://www.phoronix.com/scan.php?pa...80-linux&num=2

                Polaris was the point where we caught up, after maybe 8 generations of HW depending on how you count them.

                Vega was problematic since that was where we switched from legacy display code to DC, so availability of display code upstream was gated by the upstreaming effort.

                We did have a full open source stack at launch IIRC, but the display bits were only available in out of tree code for a while.

                https://www.phoronix.com/scan.php?pa...a-linux1&num=1
                Last edited by bridgman; 11 August 2021, 06:09 PM.
                Test signature

                Comment


                • #9
                  Originally posted by Qaridarium
                  FSR
                  FSR is a dumb image scaler + dumb image sharpen. It has some clever maths, but is extremely limited by definition and some TVs probably have better processing.
                  I've recently run some tests and it looked hardly better than simple bicubic scaling offered by ancient games + CAS...

                  Comment


                  • #10
                    Did anyone manage to run Doom Eternal (via Steam/Proton) with Raytracing enabled on "PRO" Vulkan drivers?

                    Comment

                    Working...
                    X