22.1b Installation methods: apt/dnf repository (recommended) vs .run file (manual)¶

🧭 Context Introduction¶

When setting up NVIDIA GPUs for AI workloads, one of the first decisions an engineer must make is how to install the NVIDIA Data Center GPU drivers. These drivers are the essential software layer that allows your operating system to communicate with the GPU hardware.

There are two primary installation methods: - Repository-based installation (using apt on Ubuntu/Debian or dnf on RHEL/CentOS) — this is the recommended approach. - Manual .run file installation — a more hands-on method used in specific scenarios.

This guide explains both approaches, their trade-offs, and when to use each.

⚙️ Method 1: Repository-Based Installation (apt/dnf) — Recommended¶

What is it?¶

This method uses your Linux distribution's native package manager to install NVIDIA drivers from NVIDIA's official package repository.

How it works¶

• You add NVIDIA's official repository to your system's package sources.
• You then use apt (for Debian/Ubuntu) or dnf (for RHEL/CentOS/Fedora) to install the driver.
• The package manager handles dependencies, versioning, and future updates automatically.

✅ Advantages¶

• Automatic updates — drivers stay current with system updates.
• Dependency management — the package manager resolves all required libraries.
• Clean uninstallation — removing the driver is as simple as a single command.
• Kernel compatibility — the package manager ensures driver and kernel versions match.
• Recommended by NVIDIA for most production and development environments.

❌ Disadvantages¶

• Requires internet access to the repository.
• May not offer the absolute latest driver version immediately (slight delay vs. .run file).
• Requires adding a third-party repository to your system.

🛠️ Method 2: Manual .run File Installation¶

What is it?¶

This method involves downloading a standalone executable .run file directly from NVIDIA's website and running it manually.

How it works¶

• You download the driver package (a single .run file) from NVIDIA's driver portal.
• You make the file executable and run it with root privileges.
• The installer handles the driver installation, but you must manage dependencies and updates yourself.

✅ Advantages¶

• Immediate access to the latest driver version as soon as NVIDIA releases it.
• No repository dependency — works on systems without internet access or custom repositories.
• Full control over installation options (e.g., disabling Nouveau, selecting components).
• Useful for testing new driver versions before they reach official repositories.

❌ Disadvantages¶

• Manual updates — you must manually download and reinstall for each driver update.
• No automatic dependency resolution — you must ensure all prerequisites (kernel headers, compilers, etc.) are installed beforehand.
• Kernel update breaks — after a kernel update, the driver may need to be reinstalled.
• More complex uninstallation — requires running the .run file with an uninstall flag.
• Not recommended for production environments or beginners.

📊 Visual Representation: NVIDIA Driver Installation flow¶

This flowchart outlines driver installation: disabling Nouveau drivers, building kernel modules, and verifying with nvidia-smi.

📊 Comparison Table: apt/dnf Repository vs .run File¶

🕵️ When to Use Each Method¶

✅ Use apt/dnf repository when:¶

• You are setting up a production system or development workstation.
• You want automatic updates and minimal maintenance.
• You have internet access to the NVIDIA repository.
• You are a new engineer learning AI infrastructure.

✅ Use .run file when:¶

• You need the absolute latest driver for testing a new GPU feature.
• Your system has no internet access (air-gapped environment).
• You need fine-grained control over installation options.
• You are troubleshooting a driver issue and need a specific version.

🧪 Practical Guidance for New Engineers¶

Starting Point¶

As a new engineer, always start with the repository-based method. It is the safest, most maintainable, and most widely supported approach.

Verification¶

After installation, verify the driver is working by checking the GPU status. You can do this by running a command that queries the driver version and GPU state.

Troubleshooting¶

• If the repository method fails, check that you have added the correct repository for your OS version.
• If the .run file method fails, ensure you have installed the required kernel headers and development tools first.
• Always reboot after driver installation to load the new kernel module.

✅ Summary¶

• Repository-based installation (apt/dnf) is the recommended method for most engineers. It is simpler, safer, and more maintainable.
• Manual .run file installation is a fallback method for specific scenarios like offline systems or bleeding-edge testing.
• For production AI infrastructure, always prefer the repository method unless you have a clear reason to use the .run file.
• As you gain experience, understanding both methods will help you handle diverse deployment environments.