Proxmox GPU Passthrough Guide (NVIDIA Tesla P4)#

1. Proxmox Host Configuration#

Pre-requisites#

Ensure PCIe Bifurcation is set to OFF in the server BIOS.

Enable IOMMU#

Open the GRUB configuration:

nano /etc/default/grub

Modify the line GRUB_CMDLINE_LINUX_DEFAULT="quiet" to:

GRUB_CMDLINE_LINUX_DEFAULT="quiet intel_iommu=on"

Update GRUB:

update-grub

Reboot the host.

Verify IOMMU Status#

dmesg | grep -e DMAR -e IOMMU

Note: Look for the line: DMAR: IOMMU enabled.

2. VM Configuration (Ubuntu 24.04)#

VM Hardware Settings#

Machine: q35
BIOS: SeaBIOS
Note: Do NOT run apt update or upgrade yet.

Add PCIe Device#

Shut down the VM.
Go to Hardware > Add > PCI Device.
Select Raw Device > Select your GPU.
Ensure All Functions and PCI-Express are checked.

3. Driver Installation in VM#

Verify Hardware Detection#

lspci -v

lspci -nnk | grep -iA2 nvidia

Prepare Build Environment#

Required for version 550+ drivers:

apt install build-essential -y

# Verify version
gcc --version

# Create GCC 13 symlink (Required fix)
ln -s -f /usr/bin/gcc-13 /usr/bin/gcc

reboot

Install NVIDIA Driver 550#

Note: Installation may pause at 81% for several minutes. This is normal.

apt install nvidia-driver-550 -y

reboot

4. Maintenance & Upgrades#

Upgrade Driver from 470 to 550#

apt-get purge nvidia-driver-470 -y

ln -s -f /usr/bin/gcc-13 /usr/bin/gcc

reboot

apt install nvidia-driver-550 -y

Monitoring Tools#

# Check version
cat /proc/driver/nvidia/version

# Standard status
nvidia-smi

# Visual monitor
apt install nvtop
nvtop

# High-frequency watch (0.5s)
watch -n 0.5 nvidia-smi

5. Performance & Reliability#

Enable Persistence Mode#

Eliminates “Cold Start” lag by keeping the driver loaded.

Create the override directory:

mkdir -p /etc/systemd/system/nvidia-persistenced.service.d

Create the override file:

cat <<EOF > /etc/systemd/system/nvidia-persistenced.service.d/override.conf
[Service]
ExecStart=
ExecStart=/usr/bin/nvidia-persistenced --user nvidia-persistenced --persistence-mode
EOF

Apply settings:

systemctl daemon-reload

systemctl restart nvidia-persistenced

Verify mode is enabled:

nvidia-smi -q | grep "Persistence Mode"

6. Docker GPU Integration#

Install NVIDIA Container Toolkit#

curl -fsSL [https://nvidia.github.io/libnvidia-container/gpgkey](https://nvidia.github.io/libnvidia-container/gpgkey) | sudo gpg --dearmor -o /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg

curl -s -L [https://nvidia.github.io/libnvidia-container/stable/deb/nvidia-container-toolkit.list](https://nvidia.github.io/libnvidia-container/stable/deb/nvidia-container-toolkit.list) | \
  sed 's#deb https://#deb [signed-by=/usr/share/keyrings/nvidia-container-toolkit-keyring.gpg] https://#g' | \
  sudo tee /etc/apt/sources.list.d/nvidia-container-toolkit.list

apt-get update && apt-get install -y nvidia-container-toolkit

Configure Docker Runtime#

nvidia-ctk runtime configure --runtime=docker

sudo systemctl restart docker