NVIDIA-CUDA-EXPORTER

Github Page Link

Overview

This README provides comprehensive information about the project, including its purpose, usage instructions, and configuration details.

Dockerized Python Application with NVIDIA CUDA Support

Dockerfile

The Dockerfile sets up a Docker image for running a Python application that utilizes NVIDIA CUDA for GPU computation. It includes the following steps:

• Starts from the NVIDIA CUDA image with Python support (nvidia/cuda:12.3.2-cudnn9-runtime-ubuntu20.04).
• Installs Python, pip, and virtualenv.
• Creates a virtual environment and activates it.
• Upgrades pip in the virtual environment.
• Copies the requirements.txt file into the container and installs dependencies.
• Makes sure the appuser owns the application directory.
• Copies the gpu_metrics.py script into the container.
• Exposes port 8888.
• Specifies the command to run the script (CMD ["python3", "gpu_metrics.py"]).

Usage

To use the Docker image:

1. Clone or download the repository containing the Dockerfile and other necessary files.
2. Build the Docker image:

docker build -t nvidia-cuda-exporter .

4. Run a Docker container based on the image:

docker run --name nvidia-cuda-exporter --gpus all -p 8888:8888 -v /usr/local/nvidia:/usr/local/nvidia nvidia-cuda-exporter

6. Access the Python application through a web browser or programmatically: http://localhost:<host-port>

Replace <host-port> with the desired port on your host machine.

NVIDIA GPU Metrics Collector

Python Script

The gpu_metrics_collector.py script collects GPU metrics using NVIDIA Management Library (NVML) and exposes them as Prometheus metrics via an HTTP server. It includes the following functionality:

• Collects various GPU metrics such as memory usage, utilization, temperature, power usage, fan speed, clock speeds, PCIe transmission speed, and more.
• Exposes the metrics at http://localhost:8888/metrics.

Usage

To use the script:

1. Clone or download the repository containing the script.
2. Install the required dependencies: pip install prometheus_client pynvml
3. Run the script: python gpu_metrics_collector.py

Access the metrics at http://localhost:8888/metrics.