External GPU (eGPU) + NVIDIA Drivers on Linux: Solving the Display Manager Initialization Problem

TL;DR: If your NVIDIA eGPU works in recovery mode but gives a black screen on normal boot, you're missing one critical Xorg option: AllowExternalGpus. This guide shows how to fix it properly on any X11-based Linux distribution.

Introduction

Installing NVIDIA drivers on a Linux system with an external GPU (eGPU) connected via Thunderbolt can result in a frustrating black screen instead of your login screen. This issue affects LightDM, SDDM, GDM (X11 session), and other display managers across multiple distributions.

This guide documents a complete solution tested on real hardware and explains the root cause that official documentation often omits.

Tested Configuration:

• Hardware: GEEKOM GT1 Mega (Intel Core Ultra 9 185H with Intel Arc iGPU) + NVIDIA RTX 5060 Ti in Sonnet eGPU Breakaway Box 750ex
• Connection: Thunderbolt 4
• OS: Linux Mint 22.3 MATE (applicable to Ubuntu, Fedora, Arch, and any X11-based distribution)
• Driver: NVIDIA 595 (proprietary)

Symptoms

Before diving into the solution, confirm you're experiencing this specific issue:

• ✅ NVIDIA drivers installed successfully
• ✅ nvidia-smi works and shows your GPU
• ✅ GPU visible in lspci output
• ❌ Black screen instead of login screen on normal boot
• ✅ System works normally in recovery mode or without X11
• ⚠️ Possible error in dmesg: i915: failed to get ACT after 3000ms
• ❌ Problem persists across different display managers (LightDM, SDDM, GDM in X11 mode)

Root Cause

NVIDIA drivers intentionally disable external GPUs by default as a safety measure to prevent crashes when the Thunderbolt cable is accidentally disconnected.

Without the AllowExternalGpus flag, the X11 server attempts to initialize the NVIDIA GPU, receives a denial, and crashes:

(WW) NVIDIA(GPU-0): This device is an external GPU, but external GPUs have not
(WW) NVIDIA(GPU-0):     been enabled with AllowExternalGpus. Disabling this device
(EE) NVIDIA(0): Failing initialization of X screen
(EE) no screens found
Fatal server error: no screens found

X11 then attempts to fall back to the Intel iGPU (modesetting driver), but if your monitor is connected only to the eGPU, there are no screens available on the Intel outputs, resulting in a black screen.

Why GNOME/Wayland might work without this fix:

Wayland bypasses X11 and interacts directly with GPUs via KMS (kernel modesetting). NVIDIA drivers don't block KMS access for eGPUs. Display managers using Wayland (like GDM in Wayland mode) will work, while X11-based sessions (LightDM, SDDM, Cinnamon, MATE) will fail.

Additional Issue: Boot Race Condition

Even after adding AllowExternalGpus, you might experience intermittent black screens. This occurs due to timing issues:

1. Display manager starts → attempts to launch X11
2. nvidia-drm module hasn't completed initialization (~2–3 seconds)
3. Thunderbolt DisplayPort tunnel establishes even later

This is addressed through systemd service synchronization (detailed in Step 4 below).

Step 1: Diagnosis

Check Xorg logs from recovery mode or TTY (Ctrl+Alt+F2):

grep -E "(EE|WW|AllowExternal|no screens|nvidia)" /var/log/Xorg.0.log

If you see references to AllowExternalGpus or no screens found, you're in the right place.

Verify GPU is visible to the system:

nvidia-smi
# Should show GPU with temperature, memory usage, etc.

lspci | grep -i nvidia
# Should list your GPU

Confirm monitor connection to eGPU:

ls /sys/class/drm/
# Look for card0-DP-* or card0-HDMI-* entries
cat /sys/class/drm/card0-DP-1/status
# Should return: connected

Step 2: The Critical Fix – AllowExternalGpus

Create or edit the X11 configuration file:

sudo nano /etc/X11/xorg.conf.d/10-nvidia.conf

File contents:

Section "ServerLayout"
    Identifier "layout"
    Screen 0 "nvidia"
    Inactive "intel"
EndSection

Section "Device"
    Identifier "nvidia"
    Driver "nvidia"
    Option "PrimaryGPU" "yes"
    Option "AllowExternalGpus" "True"
EndSection

Section "Screen"
    Identifier "nvidia"
    Device "nvidia"
    Option "AllowEmptyInitialConfiguration"
EndSection

Section "Device"
    Identifier "intel"
    Driver "modesetting"
EndSection

Section "Screen"
    Identifier "intel"
    Device "intel"
EndSection

Critical line: Option "AllowExternalGpus" "True" — nothing works without this.

Option "AllowEmptyInitialConfiguration" — allows X11 to start even if the GPU isn't fully initialized when the display manager launches.

Step 3: Kernel Mode Setting (KMS) Configuration

If not already configured during driver installation:

# Verify modeset is enabled
cat /proc/cmdline | grep nvidia-drm
# Should show: nvidia-drm.modeset=1

If missing, add to GRUB:

sudo nano /etc/default/grub

Locate GRUB_CMDLINE_LINUX_DEFAULT and add parameters:

GRUB_CMDLINE_LINUX_DEFAULT="quiet splash nvidia-drm.modeset=1"

Create modprobe configuration:

sudo nano /etc/modprobe.d/nvidia-kms.conf

options nvidia-drm modeset=1
options nvidia NVreg_PreserveVideoMemoryAllocations=1

Apply changes:

sudo update-grub
sudo update-initramfs -u

Step 4: Boot Race Condition Fix (for stability)

This is optional but eliminates rare black screens on some boots.

GPU Wait Script

sudo nano /usr/local/bin/nvidia-egpu-wait.sh

#!/bin/bash
# Wait for NVIDIA GPU to appear in /sys/class/drm
TIMEOUT=30
COUNT=0
while [ $COUNT -lt $TIMEOUT ]; do
    if ls /sys/class/drm/ 2>/dev/null | grep -q "^card[0-9]$"; then
        # Verify it's NVIDIA, not just Intel
        for card in /sys/class/drm/card[0-9]; do
            vendor=$(cat "$card/device/vendor" 2>/dev/null)
            if [ "$vendor" = "0x10de" ]; then
                sleep 2  # Additional pause for TB3 DP tunnel
                exit 0
            fi
        done
    fi
    sleep 1
    COUNT=$((COUNT + 1))
done
exit 0  # Timeout - continue anyway

sudo chmod +x /usr/local/bin/nvidia-egpu-wait.sh

Hotplug Script (runs after display manager)

sudo nano /usr/local/bin/nvidia-drm-hotplug.sh

#!/bin/bash
sleep 8
udevadm trigger --action=change --subsystem-match=drm
udevadm settle

sudo chmod +x /usr/local/bin/nvidia-drm-hotplug.sh

Systemd Service: Wait (runs BEFORE display manager)

sudo nano /etc/systemd/system/nvidia-egpu-wait.service

[Unit]
Description=Wait for NVIDIA eGPU initialization
After=bolt.service
Before=display-manager.service
DefaultDependencies=no

[Service]
Type=oneshot
ExecStart=/usr/local/bin/nvidia-egpu-wait.sh
RemainAfterExit=yes
TimeoutSec=35

[Install]
WantedBy=display-manager.service

Systemd Service: Hotplug (runs AFTER display manager)

sudo nano /etc/systemd/system/nvidia-drm-hotplug.service

[Unit]
Description=NVIDIA DRM hotplug trigger after display manager
After=display-manager.service bolt.service
Wants=display-manager.service

[Service]
Type=oneshot
ExecStart=/usr/local/bin/nvidia-drm-hotplug.sh
RemainAfterExit=no

[Install]
WantedBy=multi-user.target

Display Manager Drop-in (LightDM example)

sudo mkdir -p /etc/systemd/system/lightdm.service.d/
sudo nano /etc/systemd/system/lightdm.service.d/wait-nvidia-egpu.conf

[Unit]
Wants=nvidia-egpu-wait.service
After=nvidia-egpu-wait.service

For SDDM, use /etc/systemd/system/sddm.service.d/ instead.

Enable Services

sudo systemctl daemon-reload
sudo systemctl enable nvidia-egpu-wait.service
sudo systemctl enable nvidia-drm-hotplug.service

Step 5: PRIME Configuration (GPU priority)

On systems with NVIDIA drivers and nvidia-prime:

sudo prime-select nvidia

Verify:

prime-select query
# Should return: nvidia

Step 6: Reboot and Verification

sudo reboot

Post-boot verification:

# GPU is active and in use
nvidia-smi

# Xorg has no critical errors
grep -E "^(EE|WW)" /var/log/Xorg.0.log

# Services completed successfully
systemctl status nvidia-egpu-wait.service
systemctl status nvidia-drm-hotplug.service

# NVIDIA is managing the display (not Intel fallback)
xrandr --listproviders
# Should show: NVIDIA-0 as primary provider

Troubleshooting: If Black Screen Persists

Boot into recovery mode → drop to root shell → check logs:

# Main X11 log
cat /var/log/Xorg.0.log | grep -E "(EE|WW|AllowExternal|screen)"

# Boot journal
journalctl -b 0 -p err --no-pager | tail -50

# Service status
systemctl status lightdm nvidia-egpu-wait nvidia-drm-hotplug

# Initialization sequence
journalctl -b 0 --no-pager | grep -E "(nvidia|drm|lightdm|sddm|bolt)" | head -40

Common Errors and Solutions

Rollback (if something goes wrong)

From recovery mode or another system (chroot):

# Remove our xorg config - X11 reverts to auto-detection
sudo rm /etc/X11/xorg.conf.d/10-nvidia.conf

# Or temporarily rename for testing
sudo mv /etc/X11/xorg.conf.d/10-nvidia.conf /etc/X11/xorg.conf.d/10-nvidia.conf.bak

Chroot from another system:

sudo mkdir -p /mnt/target
sudo mount /dev/nvme0n1pX /mnt/target  # replace X with your partition
sudo mount --bind /dev /mnt/target/dev
sudo mount --bind /proc /mnt/target/proc
sudo mount --bind /sys /mnt/target/sys
sudo mount --bind /run /mnt/target/run
sudo chroot /mnt/target /bin/bash
# Make changes...
exit
sudo umount /mnt/target/dev /mnt/target/proc /mnt/target/sys /mnt/target/run
sudo umount /mnt/target

Final Configuration Summary

Files that should exist after setup:

/etc/X11/xorg.conf.d/10-nvidia.conf          ← primary fix
/etc/modprobe.d/nvidia-kms.conf               ← KMS modeset
/etc/default/grub                             ← nvidia-drm.modeset=1 in cmdline
/usr/local/bin/nvidia-egpu-wait.sh            ← wait script
/usr/local/bin/nvidia-drm-hotplug.sh          ← hotplug script
/etc/systemd/system/nvidia-egpu-wait.service  ← service (Before=DM)
/etc/systemd/system/nvidia-drm-hotplug.service← service (After=DM)
/etc/systemd/system/lightdm.service.d/wait-nvidia-egpu.conf  ← drop-in

Key kernel parameters (in /proc/cmdline):

nvidia-drm.modeset=1

Technical Explanations

Why the problem isn't the display manager:

LightDM, SDDM, and GDM are just wrappers that launch X11. They all use the same X server (/usr/bin/Xorg). The root cause lies in NVIDIA driver behavior at the X11 level, not in the display manager itself.

Why GNOME/Wayland worked without the fix:

GNOME defaults to Wayland, which interacts with GPUs via KMS (kernel modesetting) directly, bypassing Xorg. NVIDIA drivers don't block KMS access for eGPUs. Therefore, GDM in Wayland mode worked while LightDM/SSDM (X11) didn't.

Why i915 ACT error is not the cause:

The Intel iGPU sees that X11 is attempting to use it as a fallback (after NVIDIA rejection) and begins initializing Intel DisplayPort outputs, but the monitor isn't connected to Intel → timeout. This is a consequence of X11 failing with NVIDIA, not the root cause.

About Thunderbolt and bolt:

If the eGPU isn't authorized in bolt, it won't appear in the system at all. Check: boltctl list. If status isn't authorized, run: sudo boltctl enroll --policy auto <uuid>.

Known Behavior: Boot Delay (30-90 seconds)

On cold boots with eGPU via Thunderbolt, you may experience a delay before the login screen appears. This is normal and relates to sequential initialization:

1. Thunderbolt authorization (~15 sec)
2. NVIDIA driver loading (~20 sec)
3. DisplayPort tunnel establishment (~15 sec)
4. X11 initialization (~10 sec)

Total: 30-60 seconds on modern hardware.

The systemd services (nvidia-egpu-wait and nvidia-drm-hotplug) minimize this delay but can't eliminate it entirely due to Thunderbolt physics.

Possible optimizations:

• Configure bolt with auto-enroll policy
• Use nvidia-smi -pm 1 for early GPU "warm-up"
• Disable unused systemd services

Conclusion

The root cause of black screen issues when using NVIDIA eGPU on Linux isn't the display manager, PRIME configuration, or GRUB parameters. It's a single missing Xorg option: AllowExternalGpus.

NVIDIA drivers disable external GPUs by default as a safety measure. Without explicit permission via this flag, X11 initialization fails silently, resulting in a black screen.

This configuration has been tested extensively and works reliably across multiple distributions. If you're building a Linux workstation with eGPU, this guide can save you hours of troubleshooting.

What we learned:

• ✅ External GPUs require explicit enablement in Xorg configuration
• ✅ Display managers (LightDM, SSDM, GDM in X11) all experience the same issue
• ✅ Wayland sessions work because they bypass X11 entirely
• ✅ Boot timing issues can be addressed with systemd service synchronization
• ✅ The i915 ACT error is a red herring — a consequence, not the cause

Questions? Feel free to ask in the comments. I'll be monitoring this thread and happy to help troubleshoot your specific configuration.

Author: Aleksandr Kossarev

Location: Estonia

Date: May 2, 2026

Hardware: GEEKOM GT1 Mega + NVIDIA RTX 5060 Ti (eGPU via Thunderbolt 4)

Project: Arche Iscrin — AI-assisted creative projects

This article is based on real-world troubleshooting and testing. All commands and configurations have been verified on actual hardware. Feel free to share this guide with anyone struggling with eGPU setup on Linux.