@Forza I spent a few more minutes on the second linked article and while it doesn't actually spell it out, it does say to use XCP-ng Center to make it easier.

I went to XCP-ng Center, went to properties on the host and then went to GPU and said "This server will not use the integrated GPU" which will take affect after reboot.

After reboot:

I can now added this to a VM:

We can go ahead and mark this as solved, I'm not sure why it didn't work as expected in the command line so maybe there could be some help testing that out.

@xenoNfluX @olivierlambert

I would agree this would be great for the Z-wave/Zigbee dongles in Home Automation.

Right now I'm working on a small homelab that can pass through the intel Quicksync to a Plex server and the USB to support HA.

Having this in XOA/XOCE, even XO-lite (understandably not really a "lite" feature).

@olivierlambert I know it's not a part of this ticket but I'm not able to export ova (vmdk) files and get them to import into anything else. I've tried vmware workstation pro, vmware fusion, and vituralbox.

I get an OVA or if I download the vmdk from the disk menu but even if I added it to an existing vm they either can't read it or it's corrupt and can't be repaired.

These VMs boot and work fine on XCP-ng (I haven't tried to reimport an exported one).

@Forza @olivierlambert

I would like to highlight that this does work but there are some performance issues with the iGPU. I it for transcoding plex and it will transcode a 4K to 1080p which is a huge load on baremetal but when I use the passthrough to the same basic Windows install it says hw transcoding. However, there is a performance hit somewhere in the passthrough as I can't do a 4K transcode.

FYI

@Darkbeldin
Seems like the slave still has it registered. When I try to add the slave to XCP-ng Center, it says it's still a part of a pool and I need to connect to that. The master just has no record and can't add the slave back because it's already connected.

I change interfaces on the pool master. There was an eth0 vs eth1 issue which was causing issues in the pool. I am SSH'd into both hosts but the slave thinks it's in the pool but the master doesn't. xe host-list on the master shows only itself. I keep getting, The uuid you supplied was invalid. on the slave no matter the command. I would just install the slave but.. it has a running VM that I would like to migrate but XO can't see.

Update: Because I have the vm on a shared SR, I was able to shutdown the vm and migrate it over manually. I think maybe due to an issue I clicked forget host. Is there a way to factory reset the host without a full reinstall?

I setup a Windows 10 VM, changed the passthrough and installed plex server and it's able to do a hardware transcoding. So it looks like the passthrough of the gpu is working but for some reason debian through a docker isn't seeing it properly.

I'm using XCP-ng with Intel GPU passthrough to a VM that is running a Plex docker. I feel I'm pretty close to having it work, I have seen others get it working but they aren't responding to emails on their blogs. Would love to get some help with this.

You can see from the post below that I was able to get passthrough setup:
Intel GPU passthrough using XCP-ng Center

In my Debian VM I can see the Intel GPU:

00:05.0 VGA compatible controller: Intel Corporation CometLake-S GT2 [UHD Graphics 630]

I can also see /dev/dri:

root@docker-2:/var/lib/docker/volumes# ls -al /dev/dri
total 0
drwxrwxrwx  3 root root        120 Sep  9 16:42 .
drwxr-xr-x 18 root root       3320 Sep 10 13:39 ..
drwxrwxrwx  2 root root        100 Sep  9 16:42 by-path
crwxrwxrwx  1 root video  226,   0 Sep  9 16:42 card0
crwxrwxrwx  1 root video  226,   1 Sep  9 16:42 card1
crwxrwxrwx  1 root render 226, 128 Sep  9 16:42 renderD128

I have plex running in docker with dri passthrough:

    devices:
      - /dev/dri:/dev/dri

Where I'm struggling is getting plex to use the hardware acceleration, I have it enabled in settings transcoder too.

Sep 10, 2022 14:28:50.687 [0x7f2aec8e2b00] Debug — [Req#17b2/Transcode] Codecs: hardware transcoding: testing API vaapi
Sep 10, 2022 14:28:50.693 [0x7f2aec8e2b00] Debug — [Req#17b2/Transcode] Codecs: hardware transcoding: testing API vaapi
Sep 10, 2022 14:28:50.854 [0x7f2aec8e2b00] Debug — [Req#1991/Transcode] TPU: hardware transcoding: using hardware decode accelerator vaapi
Sep 10, 2022 14:28:50.854 [0x7f2aec8e2b00] Debug — [Req#1991/Transcode] TPU: hardware transcoding: zero-copy support present
Sep 10, 2022 14:28:50.854 [0x7f2aec8e2b00] Debug — [Req#1991/Transcode] TPU: hardware transcoding: using zero-copy transcoding
Sep 10, 2022 14:28:50.854 [0x7f2aec8e2b00] Debug — [Req#1991/Transcode] Codecs: hardware transcoding: testing API vaapi
Sep 10, 2022 14:28:50.856 [0x7f2aec8e2b00] Debug — [Req#1991/Transcode] TPU: hardware transcoding: final decoder: vaapi, final encoder: vaapi
Sep 10, 2022 14:28:51.130 [0x7f2aec6d4b00] Debug — TPU: hardware transcoding: enabled, but no hardware decode accelerator found
Sep 10, 2022 14:28:51.130 [0x7f2aec6d4b00] Debug — Codecs: hardware transcoding: testing API vaapi
Sep 10, 2022 14:28:51.132 [0x7f2aec6d4b00] Debug — TPU: hardware transcoding: final decoder: , final encoder: 
Sep 10, 2022 14:29:01.362 [0x7f2aec8e2b00] Debug — [Req#1b4a/Transcode/64c67i60wqb62oh7tdwatfv7] TPU: hardware transcoding: enabled, but no hardware decode accelerator found
Sep 10, 2022 14:29:01.363 [0x7f2aec8e2b00] Debug — [Req#1b4a/Transcode/64c67i60wqb62oh7tdwatfv7] Codecs: hardware transcoding: testing API vaapi
Sep 10, 2022 14:29:01.372 [0x7f2aec8e2b00] Debug — [Req#1b4a/Transcode/64c67i60wqb62oh7tdwatfv7] TPU: hardware transcoding: final decoder: , final encoder: 

@olivierlambert
Thanks, feel free to mark this as solved.

When I use cloud-init to build a new vm, the 10MB config drive is still there. Can I get rid of that or does that reconfigure the vm at every boot?

@xenoNfluX @olivierlambert

I would agree this would be great for the Z-wave/Zigbee dongles in Home Automation.

Right now I'm working on a small homelab that can pass through the intel Quicksync to a Plex server and the USB to support HA.

Having this in XOA/XOCE, even XO-lite (understandably not really a "lite" feature).

@Forza I spent a few more minutes on the second linked article and while it doesn't actually spell it out, it does say to use XCP-ng Center to make it easier.

I went to XCP-ng Center, went to properties on the host and then went to GPU and said "This server will not use the integrated GPU" which will take affect after reboot.

After reboot:

I can now added this to a VM:

We can go ahead and mark this as solved, I'm not sure why it didn't work as expected in the command line so maybe there could be some help testing that out.