Announcement

Collapse
No announcement yet.

AMD/Ryzen NPT Fix Discovered For Better Pass-Through Graphics Performance

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

  • AMD/Ryzen NPT Fix Discovered For Better Pass-Through Graphics Performance

    Phoronix: AMD/Ryzen NPT Fix Discovered For Better Pass-Through Graphics Performance

    One area where AMD Ryzen users have encountered Linux issues with virtualization is when trying to setup pass-through support for a graphics card to allow the virtual machine direct access to the GPU. When NPT (Nested Page Tables) are enabled, performance can become severely degraded...

    http://www.phoronix.com/scan.php?pag...U-Pass-Through

  • #2
    BTW in subsequent emails from that thread there is some debate about what a proper fix would look like, but this looks like enough to let people start working out what that proper fix would be. Another user (Nick Sarnie) passed the word to us a couple of days ago and our server SW team is going to have someone work on a "hopefully final" fix.

    One interesting thing I found was that before kernel 4.2 the value that Geoff changed was hard-coded very similarly to what he did, but between 4.1 and 4.2 the hard-coding was removed. There was apparently some debate at the time re: whether that change was correct.

    Nice work by Geoff for sure.
    Last edited by bridgman; 10-27-2017, 04:31 PM.

    Comment


    • #3
      this kind of detective work is what makes Open Source great, and so much better in the long term, sustainable future.

      Comment


      • #4
        Now, let us see some numbers with (current) Radeon gfks.
        Not that I'm owning a Ryzen based system, yet... --- Anyone? ;-)

        GREAT finding and much appreciated for _server_ workloads.

        Comment


        • #5
          Awesome news!

          Has anybody tried to pass-through an RX 480?

          I read a while back that there was some bug with pass-through in Linux that caused a VM not to work after a reboot without restarting the host itself.

          Does anybody by chance know if this is still the case with a Ryzen motherboard and a RX480?

          Comment


          • #6
            Originally posted by bridgman View Post
            BTW in subsequent emails from that thread there is some debate about what a proper fix would look like, but this looks like enough to let people start working out what that proper fix would be. Another user (Nick Sarnie) passed the word to us a couple of days ago and our server SW team is going to have someone work on a "hopefully final" fix.

            One interesting thing I found was that before kernel 4.2 the value that Nick changed was hard-coded very similarly to what he did, but between 4.1 and 4.2 the hard-coding was removed. There was apparently some debate at the time re: whether that change was correct.

            Nice work by Geoff for sure.
            It was Geoff who changed the value, I'm only the messenger

            Comment


            • #7
              I have an AMD FX CPU and an RX 480 GPU, I've tried to have GPU passthrough on a VM, but it didn't work. Could that bug be linked to my problem? Or it's simply that my setup can't handle GPU passthrough?

              Comment


              • #8
                I have a question concerning the GPU passthrough. How can I get my mouse and keyboard to interact with the VM without having a second USB mouse and keyboard? I tried using a keyboard-mouse switcher (a KVM without video) and switching back to the VM causes the keyboard and mouse to malfunction. I don't have Linux handy, but I remember that every time I unplug my mouse and keyboard, the ID number increments and I did not want that to happen. I'm at World Services for the Blind with only a Windows laptop as I'm getting certified in A+ and Network+, but my question about VM in Linux with PCIe passthrough still stands.

                Comment


                • #9
                  Originally posted by JPFSanders View Post
                  Awesome news!

                  Has anybody tried to pass-through an RX 480?

                  I read a while back that there was some bug with pass-through in Linux that caused a VM not to work after a reboot without restarting the host itself.

                  Does anybody by chance know if this is still the case with a Ryzen motherboard and a RX480?
                  I've used a 580, the GPU resets if the guest shuts down gracefully. Crashes are a tossup whether it resets properly.

                  Comment


                  • #10
                    Originally posted by GraysonPeddie View Post
                    I have a question concerning the GPU passthrough. How can I get my mouse and keyboard to interact with the VM without having a second USB mouse and keyboard? I tried using a keyboard-mouse switcher (a KVM without video) and switching back to the VM causes the keyboard and mouse to malfunction. I don't have Linux handy, but I remember that every time I unplug my mouse and keyboard, the ID number increments and I did not want that to happen. I'm at World Services for the Blind with only a Windows laptop as I'm getting certified in A+ and Network+, but my question about VM in Linux with PCIe passthrough still stands.
                    Use Synergy. Some games will have unbelievably high mouse sensitivity unless you use lock to a screen.

                    Comment

                    Working...
                    X