Quadro K2000 passthrough help

I’ve been trying for the past 4 days to get his graphics card to pass to a OVMF guest, Q35 machine. I have enabled IOMMU groups, I have turned off CSM and enabled above 4G encoding. I have followed the Proxmox Wiki on this to the letter. I have added the drivers to blacklist, I have GPU ids to vfio.conf to prevent a init event, I have used the rom-parser to make sure the GPU is UEFI capable and I’m at a wits end here.

I have tried two VM’s one running Seabios on a ubuntu 20.04 and another running OVMF on a ubuntu 20.04 the purpose of this is to run a jupyter notebook for machine learning tasks using the docker container image.

If anyone has any suggestions i will try them.
PC specs: Z10PE WS 16 running two E5-2673 v3s with 64GB or ECC ram,
its worth to mention that this is the only GPU in the machine not counting the integrated one and the operating system is Proxmox and the conf file is here:

agent: 1
bios: ovmf
args: -cpu ‘host,+kvm_pv_unhalt,+kvm_pv_eoi,hv_vendor_id=proxmox,hv_spinlocks=0x1fff,hv_vapic,hv_time,hv_reset,hv_vpindex,hv_runtime,hv_relaxed,hv_synic,hv_stimer,hv_tlbflush,hv_ipi,kvm=off’
boot: order=virtio0;ide2;net0
cores: 8
cpu: host,hidden=1,flags=+pcid
efidisk0: tank2TB:vm-113-disk-1,size=1M
hostpci0: 82:00,pcie=1,romfile=pnyk2000.rom,x-vga=1
ide2: none,media=cdrom
machine: pc-q35-3.1
memory: 4096
name: Portainer3-VM
net0: virtio=12:E1:C4:DB:E2:3B,bridge=vmbr0,firewall=1
numa: 1
ostype: l26
parent: before_nvidia
scsihw: virtio-scsi-pci
smbios1: uuid=dfe2bfee-fbbc-4650-bfc8-64cced20afd3
sockets: 1
virtio0: tank2TB:vm-113-disk-0,cache=writeback,size=64G
vmgenid: b12c2ce2-b406-43d2-875c-ab17499d6f55

UPDATE:
The card did not work in UEFI mode because the vbios in the card was not the correct one for it. Once i fixed that it worked so lesson learned on buying used stuff.

What is not working? Is it throwing an error? Or seeming to boot, but nothing comes out of the GPU?

I had this GPU passed to a Win 10 VM and it was working, but when i deleted it and installed ubuntu 20.04 the code is always “WARNING: infoROM is corrupted at gpu 0000:00:10.0” on the guest and my test kernels do not produce correct outputs. I can also boot into the guest with a shell and it works fine. With further research this appears to be Nvidia code 14

It looks like either this is a bug in the Nvidia Linux driver (does nouveau work?), or possibly a bug in KVM or something where the passthrough is not working correctly, but WIndows just doesn’t care while Linux does.

I have not checked if the open source drivers work, I have tried lots of versions for the nvidia driver from 340 and up. Shoud I buy a 10$ GPU on ebay and set that one as the primary for the hypervisor to see if that helps? I was also thinking in buying a Testla M40 if this works. The testla M40 uses Maxwell 2.0 which is reported to have a higher success rate for VFIO

I’m not convinced either will help since it works in a Windows VM.

I’ve never run into this before so :man_shrugging:, sorry.

its cool just talking to someone about this helps also maybe another poor soul runs through this post and doesnt lose 5 days of time on this. I for whatever reason am going to try to add another gpu and use a HDMI dummy plug on it and see if that helps. I have a 8490HD somewhere and will use that fingers crossed!

adding the second card did not make it work.