GPU Passthru not working properly

Apps and Virtualization
,
1

(I know that it is an old GPU but it is working properly and I don’t need more than this for what I want to do.)

-Windows 10 VM 64 bits
-8GB of RAM
-250GB of SSD space
-Using an athlon 3000g with it’s integrated GPU dedicated to TrueNAS.
-the hardware is working properly
-“NVIDIA GeForce GT 740” is showing in device manager but with an exclamation mark
-PCI Device is showing in device manager but with an exclamation mark
-PCI Simple Communications Controller is showing in device manager but with an exclamation mark
-In Truenas scale the gpu is isolated

When I try to select the gpu, the following message comes:

[EINVAL] gpu_settings.isolated_gpu_pci_ids: pci_0000_01_00_1, pci_0000_01_00_0 GPU pci slot(s) are not available or a GPU is not configured.

Error: Traceback (most recent call last):
File “/usr/lib/python3/dist-packages/middlewared/main.py”, line 198, in call_method
result = await self.middleware.call_with_audit(message[‘method’], serviceobj, methodobj, params, self)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File “/usr/lib/python3/dist-packages/middlewared/main.py”, line 1466, in call_with_audit
result = await self._call(method, serviceobj, methodobj, params, app=app,
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File “/usr/lib/python3/dist-packages/middlewared/main.py”, line 1417, in _call
return await methodobj(*prepared_call.args)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File “/usr/lib/python3/dist-packages/middlewared/schema/processor.py”, line 187, in nf
return await func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^
File “/usr/lib/python3/dist-packages/middlewared/schema/processor.py”, line 47, in nf
res = await f(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^
File “/usr/lib/python3/dist-packages/middlewared/plugins/system_advanced/gpu.py”, line 44, in update_gpu_pci_ids
verrors.check()
File “/usr/lib/python3/dist-packages/middlewared/service_exception.py”, line 70, in check
raise self
middlewared.service_exception.ValidationErrors: [EINVAL] gpu_settings.isolated_gpu_pci_ids: pci_0000_01_00_1, pci_0000_01_00_0 GPU pci slot(s) are not available or a GPU is not configured.

BIOS:

-Bios settings are ok, virtualization is on and iommu is enable

-IOMMU group is correct:

IOMMU Group 8 01:00.0 VGA compatible controller [0300]: NVIDIA Corporation GK107 [GeForce GT 740] [10de:0fc8] (rev a1)
IOMMU Group 8 01:00.1 Audio device [0403]: NVIDIA Corporation GK107 HDMI Audio Controller [10de:0e1b] (rev a1)

Does anyone know what is going on? for me, it seems like TrueNAS grabs the GPU and “loses it” when trying to dedicate it to the VM.

2

I have the same issue with an intel ARC a310 card on 24.02.2 version (AMD Epyc 3151 platform)

3

It seems to me that TrueNAS is TrueDead, read many comments here and there are many problems/bugs that are just like mine, just empty talk.

I moved from TrueNAS to Proxmox + Openmediavault and it works.

Btw, VM in Proxmox is way better, and OMV is great and simple.

Just install your server on proxmox’s VM and it just works.

4

Yes i see that, no one respond, i hope a dev could look at my ticket. I’ve invested my money on that system, i would like to not change it for this thing.

5

I will say this: even in ProxMox or XCP, passing through an Intel Arc is a problem. Because no system supports passing through Rebar, an Intel Arc 380 and the Intel Arc 310 or 770 (all of which seem to use the same media decoder so they perform the same in the test I did) they perform like a DOG in virtualized mode. I mean -bad-. The impact makes them run not much better than running on CPU in virtualized mode.

In regards to TrueNas vs. OMV, etc. TrueNas (IMHO) is definitely (in regards to Dragonfish/etc.) becoming far more oriented to several growth in items that are really relevant to the small business/enterprise world and they are kicking ass on those fronts with improved auditing, NFS and SMB share monitoring, etc.

Today (Sept. 24) should be the release of 24.10, Electric Eel, which introduces a few changes I am seriously looking forward to.

I use TrueNas inside of Proxmox for similar reasons, but I think it’s a YMMV. I had tried OMV, and OMV has some things I liked, but some of the features I need simply don’t exist there - and those same features may not be needed by others. I get it.

6

Intel Arc is new so I understand why.

I think that you can use TrueNAS inside Proxmox, no problems if it works for you.

The problem for me is the requirements for using it, TrueNAS doe not work with old hardware and it’s really a shame.

But if I had to pay for something, then I wouldn’t be using TrueNAS… I would rather buy the perpetual license of Unraid for $250.

7

The dev close my Ticket in the hope of a new kernel will resolve the issue. I think that what you say is correct for the truenas part but I didn’t see in my research the hogged performance you are talking about. Have you a link?
Btw I will try to move my Plex instance from my VM to the eel docker system to get it working

8

Thats simply not true.
PCI passthrough is tricky, and even the proxmox guys consider it experimental. It doesnt even come with support for it out of the box.

It seems a lot more ppl have problems with AMD systems with PCI passthrough, while Intel systems seem to work better. At least when using server grade hardware.

Also its called TrueNAS. Its main function is as a NAS, and not a hypervisor.

9

Yeah, so no. It isn’t ProxMox or TrueNas (I pay for licenses for both). But I think that part of it is that it is the specific features you are looking for that you need. Passthrough is a feature that is difficult for every system based on the device you are passing through and based on what you are using it on. Passing through a device on some Intel boards just doesn’t work right because of their problem with device grouping; some AMD boards do IOMMU wrong - and that can be by manufacturer.

The problem with the Arc will follow you to Unraid or to anything else. The reason is because it requires rebar to be on. When you use docker to kubernetes to run Jellyfin/Plex/etc. rebar must be disabled or doesn’t function. Without it, the speed of conversion is less than 1/2 of normal. It kills Intel’s driver base.

10

Release is scheduled for some time in October.

11

?

How so?

I mean, it may not be recommended, but how old are you talking? I have it running on systems which are over 16 years old.