Announcement

Collapse
No announcement yet.

RADV Vulkan Driver's PRIME Code Rewritten

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

  • RADV Vulkan Driver's PRIME Code Rewritten

    Phoronix: RADV Vulkan Driver's PRIME Code Rewritten

    Red Hat's David Airlie keeps to work on improving the open-source Radeon Vulkan driver...

    http://www.phoronix.com/scan.php?pag...-PRIME-Rewrite

  • #2
    I don't understand why you would ever want this? The application is meant to choice the best GPU to use as per the Vk spec.

    Comment


    • #3
      funfunctor

      The article says why it's needed:
      "David Airlie has published a rewritten version of his RADV driver patches to implement basic PRIME support for better dealing with multiple GPUs where one GPU that does the rendering may need to transfer it to another GPU for display on the same Linux system in a Vulkan world"

      Comment


      • #4
        Originally posted by funfunctor View Post
        I don't understand why you would ever want this? The application is meant to choice the best GPU to use as per the Vk spec.
        It's more straightforward when you have the same functionality implemented in several places. Keeps things from getting awkward and/or unwieldy
        Last edited by bug77; 02-21-2017, 09:00 AM.

        Comment


        • #5
          iame6162013 Maybe you should read the Vk spec before spewing stuff verbatim and your understand what I mean.

          bug77 seems very unwise to me.

          Comment


          • #6
            Originally posted by funfunctor View Post
            bug77 seems very unwise to me.
            Hence the grin at the end

            Comment


            • #7
              I think this is about two GPUs rendering alternating frames - so replace with "one GPU that does [part of] the rendering"?

              I don't understand why you would ever want this? The application is meant to choice the best GPU to use as per the Vk spec.
              With the current state of Vulkan it's very hard to detect the best setup - there just isn't enough information currently exposed by the API (unless you have access to additional information on the GPUs that you can look up via the vendor and device ID). So actually figuring out how and when to do things like alternate frame rendering, how to best use GPU memory, or even which GPU to use might end up needing to be end-user configurable if apps want to allow users to get the most out of their hardware.

              Higher-level APIs had many issues, but they did have a big advantage in that their drivers fully understood the best way to utilize their hardware.
              IMHO Vulkan needs some sort of API extension to replace that - where developers can give detailed information on how they want to use certain resources (mainly queues and memory) and get recommendations from the driver, instead of trying to guess from the small amount of information available now.

              VK_KHR_get_physical_device_properties2 will hopefully help somewhat.

              Comment


              • #8
                The way I read it, it makes perfect sense, and is just for presenting frames back to the display (or compositor, maybe, I didn't double check) when it is hooked up to a GPU that isn't the one doing the rendering (say, an nVidia GPU is doing the actual work, but the display is connected to the motherboard, not the nVidia GPU, and frames have to go trough the integrated graphics chipset).

                Please take this as a very, *very* lousy description, as I didn't do proper fact/vocabulary checking. It is just intended to help understand the article a bit, as I strongly doubt it is about mGPU or alternate frame rendering (so the most correct comment would be iame6162013's).

                I will (or I hope I will) probably do my homework and come back with a more technically accurate description later.
                Last edited by [email protected]; 02-21-2017, 09:50 AM.

                Comment


                • #9
                  Originally posted by xaxazak View Post
                  I think this is about two GPUs rendering alternating frames - so replace with "one GPU that does [part of] the rendering"?
                  Nope, this is about typical PRIME setups, where you can render on both the integrated and the dedicated GPU, but only the integrated one can present (usually Intel + NVIDIA/AMD combination), because it is the one which manages (and connected to) physical output.

                  Comment


                  • #10
                    Originally posted by yzsolt View Post

                    Nope, this is about typical PRIME setups, where you can render on both the integrated and the dedicated GPU, but only the integrated one can present (usually Intel + NVIDIA/AMD combination), because it is the one which manages (and connected to) physical output.
                    More or less how it should be. I still want to use my IGP for my desktop and activate my gpu only as needed for my games.

                    Comment

                    Working...
                    X