Hey all, cross-posting from Reddit as I'm stuck. Running into a hard crash every time I try to start a Windows VM with my RTX 5060 Ti passed through. Goal is to pass it through to a Windows 11 VM for gaming and CAD. The problem is that starting the Windows VM with the GPU passed through hard crashes the host every time. SSH and web UI both go unresponsive, though the host stays pingable. Requires a physical power cycle. No kernel panic is written to the journal. The crash is hard enough that nothing gets logged. IOMMU is definitely working. I already have the iGPU passed through to a separate Ubuntu VM without issues. The 5060 Ti and its HDMI audio are in their own clean IOMMU group with nothing else sharing it.

System:
ASUS Z890, Intel Core Ultra (Arrow Lake), 64GB DDR5, Proxmox 9.2.3. GPU is a ZOTAC RTX 5060 Ti (GB206, 16GB VRAM).

IOMMU Group 13 00:1f.0 ISA bridge [0601]: Intel Corporation Device [8086:ae0d] (rev 10)
IOMMU Group 13 00:1f.5 Serial bus controller [0c80]: Intel Corporation Arrow Lake-HX SPI (flash) Controller [8086:ae23] (rev 10)
IOMMU Group 14 01:00.0 VGA compatible controller [0300]: NVIDIA Corporation GB206 [GeForce RTX 5060 Ti] [10de:2d04] (rev a1)
IOMMU Group 14 01:00.1 Audio device [0403]: NVIDIA Corporation GB206 High Definition Audio Controller [10de:22eb] (rev a1)
IOMMU Group 15 02:00.0 Non-Volatile memory controller [0108]: Samsung Electronics Co Ltd NVMe SSD Controller S4LV008[Pascal] [144d:a80c]

Current kernel cmdline:
BOOT_IMAGE=/boot/vmlinuz-7.0.6-2-pve root=/dev/mapper/pve-root ro quiet intel_iommu=on iommu=pt vfio-pci.ids=10de:2d04,10de:22eb modprobe.blacklist=snd_hda_intel

lspci -nnk for nvidia devices (before VM start):
01:00.0 VGA compatible controller [0300]: NVIDIA Corporation GB206 [GeForce RTX 5060 Ti] [10de:2d04] (rev a1)
Subsystem: ZOTAC International (MCO) Ltd. Device [19da:1772]
Kernel driver in use: vfio-pci
Kernel modules: nvidiafb, nouveau, nova_core
01:00.1 Audio device [0403]: NVIDIA Corporation GB206 High Definition Audio Controller [10de:22eb] (rev a1)
Subsystem: NVIDIA Corporation Device [10de:0000]
Kernel driver in use: vfio-pci
Kernel modules: snd_hda_intel
02:00.0 Non-Volatile memory controller [0108]: Samsung Electronics Co Ltd NVMe SSD Controller S4LV008[Pascal] [144d:a80c]

blacklist-nvidia.conf:
blacklist nouveau
blacklist nvidia
blacklist nova_core
blacklist snd_hda_codec_hdmi

VM config (104.conf):
bios: ovmf
boot: order=scsi0;ide2;ide0;net0
cores: 8
cpu: host
efidisk0: local-lvm:vm-104-disk-0,efitype=4m,size=4M
hostpci0: 0000:01:00,pcie=1
ide0: local:iso/virtio-win.iso,media=cdrom,size=771138K
ide2: local:iso/Win11_25H2_English_x64_v2.iso,media=cdrom,size=8273050K
machine: pc-q35-11.0
memory: 16096
meta: creation-qemu=11.0.0,ctime=1780768092
name: windowsclone2
net0: virtio=BC:24:11:59:58:C9,bridge=vmbr0,firewall=1,link_down=1
numa: 0
ostype: win11
scsi0: local-lvm:vm-104-disk-1,discard=on,iothread=1,size=200G
scsihw: virtio-scsi-single
smbios1: uuid=9d5381c5-8cc2-428e-8eef-c0bc9e329afb
sockets: 1
tpmstate0: local-lvm:vm-104-disk-2,size=4M,version=v2.0
vmgenid: 8c4ef5cc-bf99-406b-bec6-f1496ff7724f

Last journal entries before crash (all crash boots show this, nothing after):
kernel: vfio-pci 0000:01:00.0: Enabling HDA controllerkernel: vfio-pci 0000:01:00.0: resettingkernel: vfio-pci 0000:01:00.0: reset donesystemd[1]: Started 104.scope(network bridge setup lines)(host unresponsive from here, nothing further logged)

Every crash boot shows no journal entries at the crash point. The host hard locks with no log. The last known kernel activity before each crash is always vfio-pci 0000:01:00.0: reset done.

Things I have tried:

• Both GPU functions confirmed bound to vfio-pci before VM start
• snd_hda_intel race condition fixed via modprobe.blacklist in cmdline (audio was being grabbed by snd_hda_intel before vfio could claim it. this was fixed but crash persists)
• iommu=pt in cmdline
• cpu type set to host
• allfunctions=1 in VM config
• Disabled Resizable BAR in BIOS (GPU was showing BAR 1 at 16GB), crash still happens
• Tried kernel 6.17.2-1-pve, still crashed
• VM boots fine with PCI device removed, confirming crash is passthrough-specific
• pcie_acs_override not tried as IOMMU groups are already cleanly isolated
• VirtIO drivers installed in the VM while GPU was not passed through

The card is confirmed working, it was previously used on a bare metal Unraid install running Ollama without issues. Has anyone gotten a 5060 Ti (GB206) working with passthrough on Proxmox, particularly on Arrow Lake / Z890? Any ideas on what could cause a hard lockup at the vfio reset stage with no log output whatsoever?

Would appreciate any help, thanks!