XCP-ng
    • Categories
    • Recent
    • Tags
    • Popular
    • Users
    • Groups
    • Register
    • Login

    Intel iGPU passthough

    Scheduled Pinned Locked Moved Hardware
    39 Posts 10 Posters 9.1k Views 11 Watching
    Loading More Posts
    • Oldest to Newest
    • Newest to Oldest
    • Most Votes
    Reply
    • Reply as topic
    Log in to reply
    This topic has been deleted. Only users with topic management privileges can see it.
    • olivierlambertO Online
      olivierlambert Vates 🪐 Co-Founder CEO
      last edited by

      Could it be PCI passthrough? That might be the easiest way

      bullerwinsB 1 Reply Last reply Reply Quote 0
      • bullerwinsB Offline
        bullerwins @olivierlambert
        last edited by

        @olivierlambert I'm not sure as it not like a discrete GPU in a dedicated PCIe slot as it's integrated into the CPU. The model I'm testing in particular is an Intel 12400.

        This is the lspci output
        b4806348-dc0e-41ca-9759-ccd8a88ad255-image.png

        It looks like there is a VGA there. Should I try this guide? https://docs.xcp-ng.org/compute/

        1 Reply Last reply Reply Quote 0
        • olivierlambertO Online
          olivierlambert Vates 🪐 Co-Founder CEO
          last edited by

          You could try to pass it through and see the result

          bullerwinsB 1 Reply Last reply Reply Quote 0
          • bullerwinsB Offline
            bullerwins @olivierlambert
            last edited by bullerwins

            @olivierlambert I tried but getting this error when turning on the VM

            INTERNAL_ERROR(xenopsd internal error: (Failure
            "Error from xenguesthelper: Populate on Demand and PCI Passthrough are mutually exclusive"))

            Not sure what it means

            EDIT: after googleing it seems that static and dynamic memory has to be the same:
            9aa61d84-dd71-4f32-9590-8cff8c365a1a-image.png

            bullerwinsB MefosheezM 3 Replies Last reply Reply Quote 2
            • bullerwinsB Offline
              bullerwins @bullerwins
              last edited by

              @bullerwins Update. The VM recognized the intel GPU:

              intel_gpu_top
              1978ef47-f998-4806-a5e0-8105a4132a2a-image.png

              lspci -k | grep -EA2 'VGA|3D'
              0e9cb348-42c8-4087-854f-3f93d022ecea-image.png

              I will test with Plex Encoding

              1 Reply Last reply Reply Quote 0
              • olivierlambertO Online
                olivierlambert Vates 🪐 Co-Founder CEO
                last edited by

                That's a good sign, please let us know!

                bullerwinsB 1 Reply Last reply Reply Quote 0
                • bullerwinsB Offline
                  bullerwins @olivierlambert
                  last edited by

                  @olivierlambert

                  I got it working!

                  Pic while transcoding with plex
                  ae4d6736-b170-4d1a-b423-ed037fb6c389-image.png

                  Plex info:
                  86dbffbe-c1e1-47bd-802a-4396e2fab7b2-image.png

                  Detected:
                  9957882a-0beb-4b5e-817b-df6a28716a7e-image.png

                  For future reference if anyone finds this with google:

                  I had to " sudo chmod -R 777 /dev/dri" inside the Ubuntu VM, otherwise it didnt work.

                  I'm using binhex plex docker image, I have to add the device to the container. It's really easy with portainer:

                  The "/dev/dri/" part is the important
                  96f5a51e-3281-4c01-aafc-803850a4ff8b-image.png

                  PS: @olivierlambert there is a typo in the xcp docs, a space is missing in Step 5:
                  https://docs.xcp-ng.org/compute/#5-put-this-pci-device-into-your-vm
                  b2d60148-73b5-4e62-8bbe-0126deab8fd5-image.png
                  This copy/pastes as:
                  xe vm-param-set other-config:pci=0/0000:04:01.0uuid=<vm uuid>

                  It should be:
                  xe vm-param-set other-config:pci=0/0000:04:01.0 uuid=<vm uuid>

                  tjkreidlT 1 Reply Last reply Reply Quote 1
                  • olivierlambertO Online
                    olivierlambert Vates 🪐 Co-Founder CEO
                    last edited by

                    Wow, great news! Happy to see you have now a working setup!!

                    Fixing the doc now 🙂

                    1 Reply Last reply Reply Quote 0
                    • olivierlambertO Online
                      olivierlambert Vates 🪐 Co-Founder CEO
                      last edited by

                      Doc is now fixed 🙂

                      1 Reply Last reply Reply Quote 0
                      • A Offline
                        adriangabura
                        last edited by

                        Could you please tell us what your motherboard is called? Last time I tried, my motherboard didn't support it, though it was Intel. Also I assume such a thing is not possible on Intel Xeon chipsets like Intel® C242 chipset?

                        bullerwinsB 1 Reply Last reply Reply Quote 0
                        • bullerwinsB Offline
                          bullerwins @adriangabura
                          last edited by

                          @adriangabura Gigabyte B760M Gaming X / DDR4 / MicroATX

                          1 Reply Last reply Reply Quote 0
                          • X Offline
                            xerxist
                            last edited by

                            Hello All,

                            Coming from ESXi 8 where I was doing this for a long time but for some reason I can't get it to work on XCP-NG 8.2.1 and 8.3 on my NUC11 and NUC13.

                            I can see the device in the VM when I do lspci, Plex also recognizes it.
                            The (hw) sign comes up but nothing happens after that.
                            Tried different OS's Ubuntu/Talos nothing seems to make it work then that its detected.

                            Are there some kernel arguments I need to pass in Grub for this to make it work?
                            Really hate to switch to something else as I'm really like the setup and how everything works out the box, but this is a show stopper for my use case. 😞

                            Hope someone can give me a hint here.

                            bullerwinsB 1 Reply Last reply Reply Quote 0
                            • bullerwinsB Offline
                              bullerwins @xerxist
                              last edited by

                              @xerxist are you using Plex in docker or native install?

                              X 1 Reply Last reply Reply Quote 0
                              • X Offline
                                xerxist @bullerwins
                                last edited by xerxist

                                @bullerwins

                                Tried docker/kubernetes Manjaro(just to switch to a later kernel)

                                Changed it to run as root.
                                Changed the permissions on the /dev/dri to 777.
                                Its just weird as it clearly sees the device.

                                4a999a7d-6686-415a-b940-5460c15326ea-image.png

                                Actually got it to repond with FFMPEG

                                ef83fdf3-1c6c-4f6a-96fb-e298e6ee504e-image.png

                                It gives an error and the GPU stays like that so it hangs, which is what happens to Plex too I guess.

                                1 Reply Last reply Reply Quote 0
                                • tjkreidlT Offline
                                  tjkreidl Ambassador @bullerwins
                                  last edited by

                                  @bullerwins Why would the driver have to be world writable (permissions 777) which seems like a security risk unless the /dev directory itself isn't. But still, a writable driver area seems very strange.

                                  X 1 Reply Last reply Reply Quote 0
                                  • X Offline
                                    xerxist @tjkreidl
                                    last edited by xerxist

                                    @tjkreidl

                                    It was to rule out any permission problems.
                                    But it seems more on the iommu side of things where it goes wrong.
                                    I’ve just tried the same on proxmox and it works right away. (Well not right away, like you need to do some grub adjustments and load modules) Not sure 🤔 what I can do to fix it. Does the xcp ng kernel take these iommu grub settings too?

                                    GRUB_CMDLINE_LINUX_DEFAULT="quiet intel_iommu=on iommu=pt"

                                    and these modules

                                    vfio
                                    vfio_iommu_type1
                                    vfio_pci
                                    vfio_virqfd

                                    tjkreidlT 1 Reply Last reply Reply Quote 0
                                    • tjkreidlT Offline
                                      tjkreidl Ambassador @xerxist
                                      last edited by

                                      @xerxist I honestly do not know, but it seems it cannot hurt to try.

                                      X 1 Reply Last reply Reply Quote 0
                                      • X Offline
                                        xerxist @tjkreidl
                                        last edited by xerxist

                                        Giving this another try 🙂

                                        @tjkreidl

                                        I couldn't find those modules so this is probably not something in xcp-ng.

                                        @bullerwins

                                        Is your vm running in EFI or BIOS?

                                        X bullerwinsB 2 Replies Last reply Reply Quote 0
                                        • X Offline
                                          xerxist @xerxist
                                          last edited by xerxist

                                          Got it working !!!!! 🙂

                                          Changed the VM from UEFI to BIOS and it started working.

                                          55120676-0c12-46cb-bdb3-7a986113f5c1-image.png

                                          Still though why would it need it be BIOS instead of UEFI

                                          Need to do some more testing as I also disabled something in the BIOS of the NUC11 for ASPM to so I'll bring up another VM with UEFI and test it again.

                                          tjkreidlT 1 Reply Last reply Reply Quote 0
                                          • tjkreidlT Offline
                                            tjkreidl Ambassador @xerxist
                                            last edited by

                                            @xerxist Good news! Some boot mechanisms will work one way or the other, or in some cases, even either one of them.
                                            It's also important to keep the BIOS up-to-date.

                                            X 1 Reply Last reply Reply Quote 0
                                            • First post
                                              Last post