How do I make a module of the latest Nvidia driver? (MiniOS 5.0)

[L65536]

I tried to make a module of the latest Nvidia driver.
By referring to this discussion: https://www.reddit.com/r/debian/comments/1jyzgrx/the_correct_way_to_install_newer_nvidia_drivers/
I used these commands to create modules under MiniOS 5.0 (Fresh Start):

wget https://developer.download.nvidia.com/compute/cuda/repos/debian12/x86_64/cuda-keyring_1.1-1_all.deb
sudo dpkg -i cuda-keyring_1.1-1_all.deb
sudo apt-get update
sudo apt2sb install linux-headers-amd64
sudo apt2sb install cuda-drivers

However the following errors occurred and no module was generated:

W: mkconf: MD subsystem is not loaded, thus I cannot scan for arrays.
W: mdadm: failed to auto-generate temporary mdadm.conf file.
umount: /run/initramfs/memory/changes/build2sb.zmBhp/union/sys: target is busy.

I also tried normal install using these commands:

wget https://developer.download.nvidia.com/compute/cuda/repos/debian12/x86_64/cuda-keyring_1.1-1_all.deb
sudo dpkg -i cuda-keyring_1.1-1_all.deb
sudo apt-get update
sudo apt install linux-headers-amd64
sudo apt install cuda-drivers

During reboot, there was error relating to nvidia-persistenced.
The boot process stopped at command line login and startx also failed.

Could anyone shed some light on this issue too?
Thanks.

[crim50n]

This is the CUDA driver for Debian 12, not the Nvidia graphics card driver for Debian 13. Are you sure you want this?

[L65536]

No, I don't have a Blackwell card. (which requires the "open" driver)
I have a Turning card (takes "open" or "proprietary" driver) and an older Pascal card ("proprietary" driver only).
I would personally prefer the "proprietary" driver but any solution to the errors stated at the beginning of the thread will be very helpful as they could be applied to other driver varieties.

Reference https://wiki.debian.org/NvidiaGraphicsDrivers

[crim50n]

Please be more specific. What AI applications are you planning to use? What graphics card do you want to configure? Do you need to output to a monitor via this graphics card, or will it only be used for AI?

It seems that you have made the task very difficult for yourself by trying to configure old graphics cards according to instructions intended for the latest generation.

[L65536]

Sorry I did not make it clear.

All I need is correctly installing any of the latest "Graphics" driver for my supported GPU using external repo or executable installer, either as plain install or as a module on MiniOS 5.0.
I used the command for [cuda-drivers] initially, because it was in the reddit's example and it includes the base "graphics" driver that I actually need, not because I want CUDA only without graphics driver.

My main graphic card, GTX 1660 Super (Turing), is still supported by both "open" and "proprietary" version of the latest Nvidia driver.
Display will be output to monitors, not headless.

One of my software project need the latest driver for the updated Nvidia Capture API version 1.9, only available on "graphics" driver version 570/580.
Extra CUDA component is actually not needed as the base graphics driver's CUDA capacity already covers most of my use cases.
I use AI for image generation (StableDiffusion) and object detection (ultralytics/pytorch/yolo) but they already work on older drivers and not really relevant to the main question.

[crim50n]

Solution Options

Since Nvidia doesn't have an official repository for Debian 13 yet, here are your options:

---

Option 1: Install via .run file (Recommended)

# 1. Download the latest driver from Nvidia
wget https://us.download.nvidia.com/XFree86/Linux-x86_64/580.27/NVIDIA-Linux-x86_64-580.27.run
chmod +x NVIDIA-Linux-x86_64-580.27.run

# 2. Install dependencies
sudo apt update
sudo apt install linux-headers-$(uname -r) build-essential dkms pkg-config libglvnd-dev

# 3. Blacklist nouveau
echo "blacklist nouveau" | sudo tee /etc/modprobe.d/blacklist-nouveau.conf
echo "options nouveau modeset=0" | sudo tee -a /etc/modprobe.d/blacklist-nouveau.conf
sudo update-initramfs -u

# 4. Reboot to text mode
sudo systemctl set-default multi-user.target
sudo reboot

# 5. After reboot, login and stop display manager
sudo systemctl stop display-manager

# 6. Run the installer
sudo ./NVIDIA-Linux-x86_64-580.27.run --dkms

# 7. After installation, reboot to graphical mode
sudo systemctl set-default graphical.target
sudo reboot

---

Option 2: Use Debian 12 repository with fix

# 1. Install the keyring
wget https://developer.download.nvidia.com/compute/cuda/repos/debian12/x86_64/cuda-keyring_1.1-1_all.deb
sudo dpkg -i cuda-keyring_1.1-1_all.deb
sudo apt update

# 2. Install the driver
sudo apt install linux-headers-$(uname -r)
sudo apt install nvidia-driver-580
# or: sudo apt install nvidia-driver-580-open

# 3. IMPORTANT: Disable nvidia-persistenced BEFORE rebooting
sudo systemctl disable nvidia-persistenced.service
sudo systemctl mask nvidia-persistenced.service

# 4. Update initramfs and reboot
sudo update-initramfs -u
sudo reboot

---

Option 3: Wait for Debian 13 repository

Monitor Nvidia's repository page for Debian 13 support:
https://developer.download.nvidia.com/compute/cuda/repos/

Once available, use the proper debian13 keyring and packages.

---

Verification

nvidia-smi
# Should show driver version 570/580 and your GTX 1660 Super

---

Note: I haven't tested these solutions personally, as I don't use Nvidia graphics cards for display output. I cannot provide further assistance beyond this information. Good luck!

[L65536]

Thank you for your patience and the comprehensive answers.
I can see clearly what steps I missed originally.
I will attempt both methods and report back here.
Thank you very much.

[L65536]

I have tested the driver installation using Option 1 and Option 2.
They both worked well except Option 2 required the following change to the package names:

sudo apt install nvidia-driver nvidia-driver-cuda
# nvidia-driver-580 was not available
# nvidia-driver-cuda was required for nvidia-smi

I also attempted to create a MiniOS module using script2sb and the following script based on Option 2:

#!/bin/bash
# sudo script2sb -s ./nvidia-driver.sh

# 1. Install the keyring
wget https://developer.download.nvidia.com/compute/cuda/repos/debian12/x86_64/cuda-keyring_1.1-1_all.deb
sudo dpkg -i cuda-keyring_1.1-1_all.deb
sudo apt update

# 2. Install the driver
sudo apt install linux-headers-$(uname -r)
sudo apt install nvidia-driver nvidia-driver-cuda

# 3. IMPORTANT: Disable nvidia-persistenced BEFORE rebooting
sudo systemctl disable nvidia-persistenced.service
sudo systemctl mask nvidia-persistenced.service

# 4. Update initramfs and reboot
sudo update-initramfs -u

However the script2sb execution ended in error with:

...
update-initramfs: Generating /boot/initrd.img-6.12.41+deb13-amd64
W: mkconf: MD subsystem is not loaded, thus I cannot scan for arrays.
W: mdadm: failed to auto-generate temporary mdadm.conf file.
Removed '/etc/systemd/system/multi-user.target.wants/nvidia-persistenced.service'.
Created symlink '/etc/systemd/system/nvidia-persistenced.service' → '/dev/null'.
update-initramfs: Generating /boot/initrd.img-6.12.41+deb13-amd64
W: mkconf: MD subsystem is not loaded, thus I cannot scan for arrays.
W: mdadm: failed to auto-generate temporary mdadm.conf file.
umount: /run/initramfs/memory/changes/build2sb.p9v6d/union/sys: target is busy.

Which was the same umount error as reported at the start of this discussion when using apt2sb.
Could you try this command in your system with Fresh Start? This does not require Nvidia hardware.

sudo apt2sb install linux-headers-amd64