Troubleshooting NVIDIA on Ubuntu 26.04: `nvidia-smi` Reports “No Devices Were Found”

Hardware and software involved

This problem occurred on the following system:

Laptop: ASUS TUF Gaming FA608PM
GPU: NVIDIA GeForce RTX 5060 Laptop GPU
GPU architecture: Blackwell
Ubuntu: Ubuntu 26.04 LTS
Kernel during diagnosis: 7.0.0-28-generic
NVIDIA driver branch: 595
Recommended package: nvidia-driver-595-open

Warning: The package and kernel version numbers in this section describe one particular incident. Future Ubuntu installations may use a different kernel and a newer NVIDIA driver branch.

Initial installation

Ubuntu recommended the open NVIDIA driver:

ubuntu-drivers devices

The relevant output was:

driver : nvidia-driver-595-open - distro non-free recommended

I installed the automatically recommended driver:

sudo ubuntu-drivers install
sudo reboot

However, after rebooting, the GPU was still unusable:

nvidia-smi

Output:

No devices were found

Running ubuntu-drivers install again did not help:

All the available drivers are already installed.

This meant that the recommended driver package was registered as installed, but the NVIDIA GPU had not initialized successfully.

Diagnostic commands

I used the following commands to inspect the running kernel, graphics devices, loaded modules, Secure Boot state, installed NVIDIA packages, and relevant kernel messages:

echo "===== Running kernel ====="
uname -r

echo "===== Graphics devices and bound drivers ====="
lspci -nnk | grep -A4 -E 'VGA|3D|Display'

echo "===== Loaded graphics modules ====="
lsmod | grep -E '^(nvidia|nouveau|amdgpu)'

echo "===== Installed NVIDIA module version ====="
modinfo -F version nvidia 2>&1

echo "===== Secure Boot ====="
mokutil --sb-state

echo "===== Installed NVIDIA packages ====="
apt list --installed 2>/dev/null \
  | grep -E 'nvidia-driver-595|linux-modules-nvidia-595|nvidia-dkms-595'

echo "===== Relevant kernel messages ====="
sudo journalctl -k -b --no-pager \
  | grep -Ei 'nvidia|NVRM|nouveau|verification|secure|RmInit|GSP|firmware|BAR|Xid' \
  | tail -n 150

The results initially looked contradictory.

The PCI device was detected correctly:

01:00.0 VGA compatible controller:
NVIDIA Corporation GB206M [GeForce RTX 5060 Max-Q / Mobile]

The kernel also appeared to have attached an NVIDIA driver:

Kernel driver in use: nvidia

The NVIDIA modules were loaded:

nvidia_uvm
nvidia_drm
nvidia_modeset
nvidia

Secure Boot was not blocking the driver:

SecureBoot disabled

Nevertheless, nvidia-smi still reported no device.

The decisive kernel error

The important error was found in the kernel log:

NVRM: The NVIDIA GPU 0000:01:00.0
NVRM: installed in this system requires use of the NVIDIA open kernel modules.

NVRM: GPU 0000:01:00.0:
RmInitAdapter failed! (0x22:0x56:1017)

NVRM: GPU 0000:01:00.0:
rm_init_adapter failed, device minor number 0

This showed that an NVIDIA kernel module had loaded, but it was the wrong flavour for this GPU.

The RTX 5060 is a Blackwell-generation GPU. It requires NVIDIA's official open kernel modules. The historical proprietary kernel-module flavour cannot initialize this GPU.

Confirming which kernel module was selected

I checked the licence and path of the nvidia module:

modinfo -F license nvidia
modinfo -n nvidia

The output was:

NVIDIA
/lib/modules/7.0.0-28-generic/kernel/nvidia-595/nvidia.ko

The licence value:

NVIDIA

identified the loaded module as the proprietary kernel-module flavour.

For the open NVIDIA kernel module, the expected licence is:

Dual MIT/GPL

The module path should also correspond to an nvidia-595-open directory.

Root cause

Both NVIDIA kernel-module flavours had been installed simultaneously.

The correct packages included:

nvidia-driver-595-open
linux-modules-nvidia-595-open-7.0.0-28-generic
linux-modules-nvidia-595-open-generic-hwe-26.04

However, the following non-open packages were also present:

linux-modules-nvidia-595-7.0.0-14-generic
linux-modules-nvidia-595-7.0.0-28-generic
linux-modules-nvidia-595-generic-hwe-26.04
linux-objects-nvidia-595-7.0.0-14-generic
linux-objects-nvidia-595-7.0.0-28-generic

Both flavours provided kernel modules with the same basic names:

nvidia.ko
nvidia-drm.ko
nvidia-modeset.ko
nvidia-uvm.ko

The kernel module index selected the proprietary copy:

/lib/modules/7.0.0-28-generic/kernel/nvidia-595/nvidia.ko

The result was:

GPU detected over PCI
        ↓
nvidia kernel module loaded
        ↓
module attached to the PCI device
        ↓
GPU initialization rejected
        ↓
nvidia-smi: No devices were found

This also explains why ubuntu-drivers install said that everything was already installed. The recommended nvidia-driver-595-open package was indeed installed, so ubuntu-drivers did not detect that another copy of nvidia.ko was being selected at runtime.

Removing the conflicting proprietary packages

Warning:

Do not run this cleanup merely because nvidia-smi fails.

First confirm that:

the GPU requires the open NVIDIA kernel modules;
both open and non-open packages are installed; and
modinfo -F license nvidia reports NVIDIA rather than Dual MIT/GPL.

The following script located installed NVIDIA 595 kernel packages whose names did not contain open:

mapfile -t badpkgs < <(
    dpkg-query -W \
      -f='${db:Status-Abbrev} ${binary:Package}\n' 2>/dev/null |
    awk '
        $1 == "ii" &&
        $2 !~ /open/ &&
        (
            $2 ~ /^linux-(modules|objects)-nvidia-595/ ||
            $2 ~ /^nvidia-(dkms|kernel-source)-595/
        ) {
            print $2
        }
    '
)

echo "Proprietary packages that will be removed:"
printf '  %s\n' "${badpkgs[@]}"

On this machine it printed:

Proprietary packages that will be removed:
  linux-modules-nvidia-595-7.0.0-14-generic
  linux-modules-nvidia-595-7.0.0-28-generic
  linux-modules-nvidia-595-generic-hwe-26.04
  linux-objects-nvidia-595-7.0.0-14-generic
  linux-objects-nvidia-595-7.0.0-28-generic

Before removing them, I checked the list carefully.

The command should not remove packages such as:

nvidia-driver-595-open
linux-modules-nvidia-595-open-...
linux-image-...
ubuntu-desktop

After confirming the package list, I removed the conflicting packages:

if ((${#badpkgs[@]})); then
    sudo apt purge "${badpkgs[@]}"
else
    echo "No installed proprietary NVIDIA 595 packages were found."
fi

Removing the stale proprietary module directory

A previous partial package removal had been insufficient because the proprietary module files were still available under the running kernel's module directory.

I stored the current kernel version:

kver=$(uname -r)

Then removed only the stale proprietary NVIDIA 595 directory:

sudo rm -rf "/lib/modules/$kver/kernel/nvidia-595"

Caution:

This path is specific to this diagnosed conflict.

Do not delete arbitrary files from /lib/modules. Confirm the selected module path with modinfo -n nvidia before removing anything.

Reinstalling only the open NVIDIA module

I then reinstalled the correct open driver packages:

sudo apt update

sudo apt install --reinstall \
    "linux-modules-nvidia-595-open-$kver" \
    linux-modules-nvidia-595-open-generic-hwe-26.04 \
    nvidia-driver-595-open

Next, I rebuilt the kernel module dependency index and the initramfs:

sudo depmod -a "$kver"
sudo update-initramfs -u -k "$kver"

Verifying the selected module before rebooting

Before restarting, I checked the module again:

modinfo -k "$kver" -F license nvidia
modinfo -k "$kver" -n nvidia

The expected result was now:

Dual MIT/GPL

and a path belonging to the open module flavour, similar to:

/lib/modules/7.0.0-28-generic/kernel/nvidia-595-open/nvidia.ko

If the result had still been:

NVIDIA

I would have searched for every remaining copy of nvidia.ko:

find "/lib/modules/$kver" -type f -name 'nvidia.ko*' \
    -print -exec dpkg -S {} \; 2>&1

This identifies both the location of each module and the package that owns it.

Rebooting and final verification

After the open module had been selected correctly:

sudo reboot

After logging in again:

nvidia-smi

The RTX 5060 was finally detected normally.

I also verified the selected module:

modinfo -F license nvidia
modinfo -n nvidia

Expected result:

Dual MIT/GPL
/lib/modules/<current-kernel>/.../nvidia-595-open/.../nvidia.ko

The external display also became usable after the NVIDIA GPU initialized correctly.

Why the `-open` driver is correct

The following three drivers should not be confused:

Driver	Description
Nouveau	Community-developed open-source Linux driver
NVIDIA proprietary kernel modules	NVIDIA's historical closed kernel-module flavour
NVIDIA open kernel modules	NVIDIA's official modern kernel-module flavour

The package used here:

nvidia-driver-595-open

is still part of NVIDIA's official driver stack. It is not Nouveau.

The word open refers mainly to the NVIDIA kernel modules. The normal NVIDIA user-space components for OpenGL, Vulkan, CUDA and device management are still used.

For this RTX 5060, the choice was not between a faster proprietary driver and a slower open driver:

Proprietary NVIDIA kernel module:
    GPU initialization fails

Open NVIDIA kernel module:
    officially supported and functional

Normal installation procedure for future systems

This conflict is not expected during every Ubuntu installation.

The normal process should remain:

sudo apt update
sudo apt full-upgrade

ubuntu-drivers devices
sudo ubuntu-drivers install

sudo reboot
nvidia-smi

For an RTX 50-series GPU, I would additionally verify:

modinfo -F license nvidia
modinfo -n nvidia

The expected licence is:

Dual MIT/GPL

The detailed purge procedure above should be treated as troubleshooting for a confirmed package conflict, not as part of every routine Ubuntu installation.

Main lesson

The presence of nvidia in lsmod or the line:

Kernel driver in use: nvidia

does not prove that the GPU initialized successfully.

A more complete verification sequence is:

nvidia-smi
modinfo -F license nvidia
modinfo -n nvidia

sudo journalctl -k -b --no-pager |
grep -Ei 'nvidia|NVRM|RmInit|GSP|firmware'

In this incident, the package manager believed that the recommended driver was installed, while the kernel was actually selecting a conflicting and unsupported copy of nvidia.ko.

Troubleshooting NVIDIA on Ubuntu 26.04: nvidia-smi Reports “No Devices Were Found”