Server rental store

GPU Drivers

# GPU Drivers

This article details the configuration and management of GPU drivers on our MediaWiki servers. Proper GPU driver configuration is crucial for tasks such as image thumbnailing, video transcoding (if enabled), and potentially future machine learning integrations. This guide is aimed at new server engineers and assumes a basic understanding of Linux server administration.

Understanding the Role of GPU Drivers

GPU drivers enable the operating system and applications like MediaWiki to communicate with the Graphics Processing Unit (GPU). In our environment, GPUs are primarily utilized to accelerate image processing tasks, reducing the load on the CPU and improving overall server responsiveness. Incorrect or outdated drivers can lead to instability, performance degradation, or even system crashes. We primarily utilize NVIDIA GPUs, so this guide will focus on NVIDIA drivers. However, the principles apply to other GPU vendors as well. See Server Hardware Overview for a complete list of server hardware.

Driver Selection and Installation

Choosing the correct driver version is critical. We generally follow a policy of using stable, tested drivers rather than the bleeding-edge releases. The specific driver version will depend on the GPU model and the kernel version in use. We primarily use the NVIDIA proprietary drivers, installed via the package manager. It's essential to avoid using Nouveau, the open-source NVIDIA driver, as it lacks the performance and features required for our workload. See Kernel Versions for details on supported kernels.

Identifying Your GPU

Before installing the driver, identify the installed GPU model. Use the following command:

```bash lspci grep -i nvidia ```

This command will output information about the NVIDIA GPU(s) present in the system. Note the model number for driver selection. For further information on system identification, refer to Server Diagnostics.

Installation Procedure

The installation process varies slightly depending on the Linux distribution. The following example demonstrates the process on Debian/Ubuntu based systems:

1. Add the NVIDIA repository to your system’s sources list. 2. Update the package list: `sudo apt update` 3. Install the recommended driver package. For example, to install driver version 535.104.05: `sudo apt install nvidia-driver-535` 4. Reboot the server: `sudo reboot`

After rebooting, verify the driver installation using `nvidia-smi`. See Troubleshooting GPU Issues for common installation problems.

Supported NVIDIA GPU Models and Recommended Drivers

The following table lists the currently supported GPU models and their recommended drivers. This table is subject to change, so always verify the latest recommendations on the Server Configuration Documentation.

GPU Model Recommended Driver Version Notes
NVIDIA Tesla T4 535.104.05 Commonly used for thumbnail generation.
NVIDIA Quadro RTX 5000 525.147.05 Used in development and testing environments.
NVIDIA GeForce RTX 3090 535.104.05 Used in Video Transcoding (if enabled). Requires specific configuration.
NVIDIA Tesla V100 470.82.00 Older model, still supported but nearing end-of-life.

Driver Configuration and Monitoring

After installation, it’s important to configure the driver for optimal performance and monitor its status.

Configuration Options

The NVIDIA driver provides a range of configuration options accessible through the `nvidia-settings` utility. Key settings include:

⚠️ *Note: All benchmark scores are approximate and may vary based on configuration. Server availability subject to stock.* ⚠️

View full article | Back to Server rental store