Announcement

Collapse
No announcement yet.

Experimenting With Virtual GPU Support On Linux 4.10 + Libvirt

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

  • Experimenting With Virtual GPU Support On Linux 4.10 + Libvirt

    Phoronix: Experimenting With Virtual GPU Support On Linux 4.10 + Libvirt

    With the Linux 4.10 kernel having initial but limited Intel Graphics Virtualization Tech support, you can begin playing with the experimental virtual GPU support using the upstream kernel and libvirt...

    http://www.phoronix.com/scan.php?pag...GPU-Experiment

  • #2
    I do hope the AMD product team just go with a "every gpu that supports it, gets it" mentality

    Comment


    • #3
      Why does everyone only ever give libvirt instructions? I want to do these things with qemu directly, so any direct command line switches would be much more helpful (people could at least post them too next to their libvirt conf)

      Comment


      • #4
        Originally posted by jhenke View Post
        Why does everyone only ever give libvirt instructions? I want to do these things with qemu directly, so any direct command line switches would be much more helpful (people could at least post them too next to their libvirt conf)
        If I had to guess, that's because QEMU is not just a virtual machine but a complete emulator. To my knowledge, Xen is based on QEMU but is compatible with libvirt. I think you get less control in terms of hardware (since it isn't a full-blown emulator) but depending what you're doing with it, that shouldn't be an issue.

        Comment


        • #5
          Originally posted by boxie View Post
          I do hope the AMD product team just go with a "every gpu that supports it, gets it" mentality
          Right. While they are still thinking about who not to give this functionality, Intel users can use it already.
          bridgman please hear us

          Comment


          • #6
            GPU pass-through is like the bat cave, if you ever manage to get it to work you have no idea how you did it and you won't be able to do it again. I spent countless hours trying to set it up and finally managed to do it. I run with it for 6-9 months or something then my hard drive failed and I lost all my settings and since then it absolutely refuse to work again no matter what I try. I'm so sick of it, it's just not worth the hassle. Of course, ALL documentation of QEMU is outdated, inaccurate and/or incomplete. I hate it!

            Comment


            • #7
              Originally posted by Staffan View Post
              GPU pass-through is like the bat cave, if you ever manage to get it to work you have no idea how you did it and you won't be able to do it again. I spent countless hours trying to set it up and finally managed to do it. I run with it for 6-9 months or something then my hard drive failed and I lost all my settings and since then it absolutely refuse to work again no matter what I try. I'm so sick of it, it's just not worth the hassle. Of course, ALL documentation of QEMU is outdated, inaccurate and/or incomplete. I hate it!
              Yes, it requires quite a lot of configuration. Using libvirt is much easier than the bare qemu. I am currently running it rock solid (except the sound part in MacOS) and planning to make an in-depth video about this topic. I agree, the documentation is not great, at all. Like -set command is mentioned, but the options or what can you set is not explained.

              Another thing is that libvirt can break your setup during updates. It happened for me. I run a non-UEFI windows 7 installation and to get my gpu working I had to use XEN as the display (I also wanted sound and libvirt can not do it without a display), because windows can not find the resources. After update, it simply did not allow me to start the VM if XEN as display was selected. Took days to get the idea to change the virtual display PCI id to make it not the default display.

              Comment


              • #8
                Originally posted by juno View Post

                Right. While they are still thinking about who not to give this functionality, Intel users can use it already.
                bridgman please hear us
                As far as I understand, the AMD solution uses sriov and this requires hardware support (you can check it with `lspci`).

                Currently, only the FirePro S series have the needed hardware bits.

                Comment


                • #9
                  Originally posted by wpoely86 View Post

                  As far as I understand, the AMD solution uses sriov and this requires hardware support (you can check it with `lspci`).

                  Currently, only the FirePro S series have the needed hardware bits.
                  This actually makes me sad.

                  But at least my passthrough setup works :/

                  Comment


                  • #10
                    Originally posted by wpoely86 View Post
                    As far as I understand, the AMD solution uses sriov and this requires hardware support (you can check it with `lspci`).

                    Currently, only the FirePro S series have the needed hardware bits.
                    Yeah, and that's obviously alright.

                    I just want upcomping products and the free driver stack having it enabled. And that will actually be a point for buying Vega. We could actually have a Zen+Vega (or successors) build that supports GPU virtualisation without having to use a 2nd GPU for the host system. How great would that be?

                    BTW: Intel also supports it for their newer processors, starting with Broadwell. But then, however, they only have this weak IGPs, nothing like AMD has, and also not this separation between pro and consumer hardware. Well, they actually have, but that's Core i vs. Xeon, but this is about the CPU side of things.
                    Last edited by juno; 01-18-2017, 07:26 PM.

                    Comment

                    Working...
                    X