Announcement

Collapse
No announcement yet.

AMDGPU Virtualization Support Updated

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

  • #11
    Originally posted by Nille View Post
    Whats why he/we ask you. For me it would be the "killer feature" why i have to buy a AMD card, but not enough for a FirePro
    Understood, but that is a product management question I can't answer (but I can ask if the product management folks are ready to make a statement this early in the development cycle).

    What I can say is "please don't make assumptions about support on consumer cards just because we are open sourcing the driver code".
    Test signature

    Comment


    • #12
      Originally posted by bridgman View Post
      What I can say is "please don't make assumptions about support on consumer cards just because we are open sourcing the driver code".
      Personally, I will not make any assumptions, but if you can tell upper management any kind of feedback, please, tell them this:

      Having GPU virtualization on consumer products would be a killer feature.
      I use an AMD GPU and always recommend it over NVIDIA. GPU virtualization on consumer cards would be one more great reason to keep buying AMD GPUs.

      Comment


      • #13
        Originally posted by Nille View Post
        Whats why he/we ask you. For me it would be the "killer feature" why i have to buy a AMD card, but not enough for a FirePro
        I second that.

        Originally posted by bridgman View Post
        Understood, but that is a product management question I can't answer (but I can ask if the product management folks are ready to make a statement this early in the development cycle).
        What I can say is "please don't make assumptions about support on consumer cards just because we are open sourcing the driver code".
        Maybe give product management a hint? A fair priced entry base without (virtual) restrictions is a key to get easily developer on board, increase acceptance further and get positive attention.

        A high-priced solution should be delimited with Support and SLAs - not by technical restrictions (imho). See Red Hat with Fedora for Developers and RHEL with SLAs for Companies.

        ... and a sold amd radeon apu/gpu is also a not sold one for someone else ...

        Comment


        • #14
          Originally posted by bridgman View Post

          Understood, but that is a product management question I can't answer (but I can ask if the product management folks are ready to make a statement this early in the development cycle).

          What I can say is "please don't make assumptions about support on consumer cards just because we are open sourcing the driver code".
          Alright, that's why we are asking. It just was not clear.

          Comment


          • #15
            Originally posted by rdeleonp View Post

            Personally, I will not make any assumptions, but if you can tell upper management any kind of feedback, please, tell them this:



            I use an AMD GPU and always recommend it over NVIDIA. GPU virtualization on consumer cards would be one more great reason to keep buying AMD GPUs.
            I think the point is this code will go into Naples and the Server space to have racks of GPGPUs to run Linux, FreeBSD, whatever that is supported and can access those racks of servers and offload compute and more [Renderfarms, etc] that all would expect to have very expensive enterprise service contracts.

            Comment


            • #16
              Originally posted by bridgman View Post

              Understood, but that is a product management question I can't answer (but I can ask if the product management folks are ready to make a statement this early in the development cycle).

              What I can say is "please don't make assumptions about support on consumer cards just because we are open sourcing the driver code".
              I'm just gonna stop by and say that IF the hardware supports it on consumer cards, and it CAN be done on consumer cards, you guys'd make me VERY happy if you implemented it on consumer cards.

              But I totally understand if that's not the case.

              Comment


              • #17
                Originally posted by Marc Driftmeyer View Post

                I think the point is this code will go into Naples and the Server space to have racks of GPGPUs to run Linux, FreeBSD, whatever that is supported and can access those racks of servers and offload compute and more [Renderfarms, etc] that all would expect to have very expensive enterprise service contracts.
                I agree, the target is clear. We can just hope that project management notices the demand in other areas:
                • Multi-seat gaming, the problem there is that FirePro is not fast enough regardless of price. Gaming GPUs are better at gaming.
                • SteamOS running "virtualised windows" to increase game support and ease moving from Windows to SteamOS (most consumers can't afford FirePro)
                Maybe Vulkan can be used for these use cases? One can just dream

                Comment


                • #18
                  So maybe you could buy Radeon Pro (eg. WX 4100, it isn't that expensive), and dedicate recourses to guest OS, which could recognize it as consumer chip (RX 460)

                  Comment


                  • #19
                    Originally posted by Jabberwocky View Post
                    the problem there is that FirePro is not fast enough regardless of price. Gaming GPUs are better at gaming.
                    Radeon GPU = FirePRO GPU
                    Config; BIOS, boards, memory differ. The pro GPU itself isn't slower. However, sometimes GPUs on PRO cards are clocked lower. That isn't necessarily true since Polaris, though.

                    BTW: I think the brand "FirePRO" is dead. "FirePRO S" is now "Radeon Instinct" (vs. Nvidia Tesla), "FirePRO W" is "Radeon Pro" (vs. Nvidia Quadro).

                    Originally posted by Space Beer View Post
                    So maybe you could buy Radeon Pro (eg. WX 4100, it isn't that expensive)
                    Not that expensive?
                    It costs 3x the amount it should, if used as consumer.
                    If you just want to have a decent GPU for your Linux and play games on virtualised Windows, you could as well buy a RX 460 and 480 for that money and have much more performance.

                    However, for some bigger consumer GPU (like maybe Vega), the feature would be amazing. No more dual-/rebooting for playing a game on Windows, no more dependency on a second GPU/APU/CPU with IGP when passing through the big GPU.
                    Last edited by juno; 11 January 2017, 09:32 PM.

                    Comment


                    • #20
                      Originally posted by juno View Post
                      No more dual-/rebooting for playing a game on Windows, no more dependency on a second GPU/APU/CPU with IGP when passing through the big GPU.
                      I'm drooling just thinking about it...

                      Comment

                      Working...
                      X