VM with GPU passthrough reboots host on shutdown

[vman81]

I installed a nvidia 1050 TI and passed it through to a windows guest, and that works great for gaming/streaming with a few cpu cores and some ram. But unfortunately it seems to reboot my physical server every time I shutdown or reboot the guest VM (since upgrading physical gpu)

I'm not seeing anything obvious in syslog, messages or kern.log

The installation did include a lot of trial and error when trying to make the nvidia passthrough work without the infamous error 43, but it seemed to work fine until I upgraded the GPU from a nvidia gt 1030 to an nvidia gtx 1050ti

the vmid.conf :

bios: ovmf
bootdisk: virtio0
cores: 12
efidisk0: local-lvm:vm-199-disk-3,size=4M
hostpci0: 04:00,pcie=1,x-vga=1
machine: q35
memory: 16384
name: gaming-ok
net0: virtio=FE:5C:F9:11:8C:0E,bridge=vmbr2
numa: 0
ostype: win10
parent: beforeboot
scsihw: virtio-scsi-pci
smbios1: uuid=367043a0-15c4-4911-b10d-7e2a2f51e273
sockets: 2
virtio0: local-lvm:vm-199-disk-2,cache=writeback,size=64G
virtio1: local-lvm:vm-199-disk-4,cache=writeback,size=40G
virtio2: satadisks:vm-199-disk-1,cache=writeback,size=512G

Any suggestions? I'm running a few uptime sensitive VMs on the host and this is really annoying.

 

[vman81]

It seems like it sometimes also just hangs the host - not a full reboot. Totally unresponsive via console

 

[Republicus]

My only success with Nvidia and GPU passthrough was after turning KVM off so later drivers could be installed.

I was using a GTX 580 and the 5 series was known to have issues reinitializing so I had to reboot my host continuously. I switched to AMD and had no such issues.