Announcement

Collapse
No announcement yet.

Mesa 17.3 With RADV Vulkan Running Great With Polaris, Starts To Outperform AMDGPU-PRO

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

  • Mesa 17.3 With RADV Vulkan Running Great With Polaris, Starts To Outperform AMDGPU-PRO

    Phoronix: Mesa 17.3 With RADV Vulkan Running Great With Polaris, Starts To Outperform AMDGPU-PRO

    Yesterday I posted some fresh benchmarks of the RX Vega between the AMDGPU+RadeonSI/RADV open-source vs. AMDGPU-PRO drivers, which showed the pure open-source driver stack performing admirably well for the latest-generation AMD graphics architecture and the community-driven RADV Vulkan driver was even performing nicely. Due to how well RADV has matured during the Mesa 17.3 cycle, here are some benchmarks using a Radeon RX 580 "Polaris" graphics card showing off its more mature support for Vulkan.

    http://www.phoronix.com/vr.php?view=25390

  • #2
    Where is Bridgman to disagree with me again on that subject, i totally say that the community makes better drivers than AMD and try to hire all those talented people like Marek cannot hide that fact.

    Comment


    • #3
      Originally posted by artivision View Post
      Where is Bridgman to disagree with me again on that subject, i totally say that the community makes better drivers than AMD and try to hire all those talented people like Marek cannot hide that fact.
      The `community' are a bunch of graduate and post doctoral candidate CS majors or other engineering disciplines who whether AMD hires them directly or not work in Silicon Valley. The `community' gives them visibility and draws the engineering hiring managers attention thus accelerating their chances of being where they already want to be: Silicon Valley. Only a fool would not hire them after the work they produce.

      Before the days of the `community' Silicon Valley had to rely on academic professors and other feeler outlets to fish for talent. It's a win/win for all parties involved.

      Comment


      • #4
        AMD is actually helping with the RADV driver, though not directly. They answer questions and have documentation, have shared utilities and headers in the code, and perhaps most of all do most of (?) LLVM development that RADV and RadeonSI both uses.

        Comment


        • #5
          Originally posted by artivision View Post
          Where is Bridgman to disagree with me again on that subject, i totally say that the community makes better drivers than AMD and try to hire all those talented people like Marek cannot hide that fact.
          Are you sure you're quoting Bridgman correctly now... ?

          Comment


          • #6
            Originally posted by ernstp View Post
            AMD is actually helping with the RADV driver, though not directly. They answer questions and have documentation, have shared utilities and headers in the code, and perhaps most of all do most of (?) LLVM development that RADV and RadeonSI both uses.
            Very true. The majority of the Vulkan driver code has been written by AMD, through LLVM, the kernel driver, and various bits of shared Mesa code. On top of that, lots of tricky hardware bugs have been found and worked around in the GL code along with the code itself providing good documentation on how to program the hardware, and the radv developers are just able to copy those fixes into their own code where necessary.

            Intel should be given some credit too. A lot of the Vulkan code in Mesa is from them, or copied from their driver and then adjusted for radv.

            Still, it really points out just how poor the internal development process at AMD is that they still haven't open sourced their existing vulkan driver when they were at the forefront of creating the API in the first place and a couple volunteer developers are able to completely implement a new driver from scratch like this.
            Last edited by smitty3268; 10-24-2017, 03:09 PM.

            Comment


            • #7
              Originally posted by smitty3268 View Post

              Very true. The majority of the Vulkan driver code has been written by AMD, through LLVM, the kernel driver, and various bits of shared Mesa code. On top of that, lots of tricky hardware bugs have been found and worked around in the GL code along with the code itself providing good documentation on how to program the hardware, and the radv developers are just able to copy those fixes into their own code where necessary.

              Intel should be given some credit too. A lot of the Vulkan code in Mesa is from them, or copied from their driver and then adjusted for radv.

              Still, it really points out just how poor the internal development process at AMD is that they still haven't open sourced their existing vulkan driver when they were at the forefront of creating the API in the first place and a couple volunteer developers are able to completely implement a new driver from scratch like this.
              radv is pretty much a great example of the open source development model at work. I realised a couple of years ago we had all these great pieces of the stack already (compiler, anv WSI, kernel bits), and it just needed some missing glue to create a vulkan driver. I actually expected AMD to open source theirs quicker and kill it off as a neat thought experiment/learning exercise, good luck to them trying to kill it now.

              Dave.

              Comment


              • #8
                Originally posted by airlied View Post

                radv is pretty much a great example of the open source development model at work. I realised a couple of years ago we had all these great pieces of the stack already (compiler, anv WSI, kernel bits), and it just needed some missing glue to create a vulkan driver. I actually expected AMD to open source theirs quicker and kill it off as a neat thought experiment/learning exercise, good luck to them trying to kill it now.

                Dave.
                AMD doesn't need to do anything. This actually makes it easier for them. They just don't opensource their own Vulkan driver and let RADV remain as the de-factor standard. A win win as they now won't have any responsibility for that code.

                Comment


                • #9
                  Possible typo?

                  Originally posted by phoronix View Post
                  While at 4K, the AMDGPU-PRO 17.3 driver was again a few frames faster
                  The graph is reversed and showing RADV taking the lead...

                  Wait, definitely a typo, since you said "17.3"...

                  Comment


                  • #10
                    fglrx devs are really bad, years of work and some guys without the same resourges destroy the performance of driver in one year?

                    Comment

                    Working...
                    X