Proxmox with nVidia for remote desktop on 3 monitors

[stratoss]

Hello,

I've been trying to set-up a desktop VM (Debian 10 with Gnome) with and without PCIe-passthrough (GTX 1030).

The end result is always a slaggy screen, especially when moving a window between screens.

Did anybody succeeded in a similar setup, or can somebody give me guidance about the settings?
Can I use the GPU to 'feed' the spice screens? If yes, where to install the drivers - on the host or in the guest?

Thanks in advance,
Stan

edit: removed the spice part of the question

 

[aaron]

Spice is it's own software GPU. I don't believe you will be able to use Spice with the passed through GPU.
Maybe another remote desktop solution will help you. X2GO comes to mind as a good candidate.

 

[stratoss]

@aaron thank you for your reply. I have removed the spice part of the question and will continue the discussion here instead of opening a new one, as the goal is the same.

In order to ensure that the GPU-PassThrough is working properly, I have connected a monitor to the 1030 GPU. However, no output is going there.

Here's what I have on the PVE:

lspci -nnk
81:00.0 VGA compatible controller [0300]: NVIDIA Corporation GP108 [10de:1d01] (rev a1)
    Subsystem: NVIDIA Corporation GP108 [GeForce GT 1030] [10de:11c7]
    Kernel driver in use: vfio-pci
    Kernel modules: nvidiafb, nouveau
81:00.1 Audio device [0403]: NVIDIA Corporation GP108 High Definition Audio Controller [10de:0fb8] (rev a1)
    Subsystem: NVIDIA Corporation GP108 High Definition Audio Controller [10de:11c7]
    Kernel driver in use: vfio-pci
    Kernel modules: snd_hda_intel

cat /etc/pve/qemu-server/103.conf
agent: 1
bootdisk: scsi0
cores: 8
hostpci0: 81:00,pcie=1,x-vga=1,romfile=nvidia-vbios.rom
ide2: local:iso/debian-10.2.0-amd64-netinst.iso,media=cdrom
machine: q35
memory: 40000
name: NewDesktop
net0: virtio=CA:2F:CB:EF:1B:B4,bridge=vmbr0,firewall=1
numa: 0
ostype: l26
scsi0: local:103/vm-103-disk-0.qcow2,size=3000G
scsihw: virtio-scsi-pci
smbios1: uuid=7d753a3b-c034-4fe7-b0c0-87f5d321d50c
sockets: 1
vga: none
vmgenid: 0e539809-3a55-4a01-867e-ad35dd37ac73

./rom-parser /usr/share/kvm/nvidia-vbios.rom
Valid ROM signature found @0h, PCIR offset 1a0h
    PCIR: type 0 (x86 PC-AT), vendor: 10de, device: 1d01, class: 030000
    PCIR: revision 0, vendor revision: 1
Valid ROM signature found @ec00h, PCIR offset 1ch
    PCIR: type 3 (EFI), vendor: 10de, device: 1d01, class: 030000
    PCIR: revision 3, vendor revision: 0
        EFI: Signature Valid, Subsystem: Boot, Machine: X64
    Last image

cat /etc/modprobe.d/blacklist.conf
blacklist radeon
blacklist nouveau
blacklist nvidia
blacklist nvidiafb

cat /etc/modprobe.d/vfio.conf
options vfio-pci ids=10de:1d01,10de:0fb8

PVE version is 6.1-5 installed on ZFS-encrypted root from a clean debian buster install.

The system boots, and stucks at about 99% 1-core CPU (12% overall) and 98% Memory usage , which happens when the Debian net-install loads and waits for my selection (attached a 'virtual' monitor to check that).

I have 2 PCI-E slots and tried to move the GPU between the two, also tried with and without the romfile, without any luck.

Am I doing something wrong, or just miss something?

Thanks in advance,
Stan

 

[stratoss]

Is it a proper expectation to receive an output on the GPU when it's passed to a VM and that VM is started with x-vga=1 ?

 

[stratoss]

Just for reference in case that somebody else is struggling too: using OVMF BIOS and setting rombar=0 fixed the matter for me. @aaron thanks for the X2Go reference, works like a charm for my goals.