TensorFlow GPU Compatibility Checker and Performance Tester

This Python script is designed to serve two main purposes:

1. GPU Compatibility Checker: It checks if TensorFlow is working correctly on your GPU and provides information about your GPU setup. It also suppresses unnecessary warnings to ensure a smooth experience.

2. Performance Tester: It demonstrates the significant performance advantage of GPU-enabled TensorFlow over the CPU by training a Convolutional Neural Network (CNN) model on the CIFAR-10 dataset for image classification. The script compares multiple metrics and provides various visualizations to help you understand the difference in performance between CPU and GPU.

Features

GPU Compatibility Check

The script performs the following tasks to check GPU compatibility:

• Suppresses warnings related to CUDA/CuDNN/TensorFlow to enhance the user experience.
• Displays information about Python, Driver, and TensorFlow versions.
• Sets the GPU as the default device if available.
• Checks for the presence of GPUs and lists local devices.

Performance Testing

The script demonstrates GPU superiority by training a CNN model on the CIFAR-10 dataset and provides various metrics and visualizations:

• Training Time Comparison: Compares training times between CPU and GPU for different batch sizes.
• Combined Metrics Comparison: Compares accuracy, recall, precision, specificity, and AUC-ROC for CPU and GPU.
• Class-Specific Accuracy: Calculates class-specific accuracy scores for each class and visualizes them.
• Class-Specific Accuracy Heatmap: Provides a heatmap visualization of class-specific accuracy scores.
• Class-Specific Accuracy Summary: Summarizes class-specific accuracy scores, ranks classes, and calculates the weighted average.

Dependencies

Before running this script, make sure you have the following prerequisites installed:

• Python
• TensorFlow
• TensorFlow Datasets
• TensorFlow Keras
• Scikit-Learn
• NumPy
• Matplotlib
• Seaborn
• tqdm
• Pandas

You can install these dependencies using pip with the following command:

pip install tensorflow tensorflow-datasets scikit-learn numpy matplotlib seaborn tqdm pandas

This youtube video leads to a docker image which is very well suited for running tensorflow with an nvidia GPU:
https://youtu.be/YozfiLI1ogY?si=Sb6q807ILbPNQiIc

You may also be able to just find the image by searching the container database:
thegeeksdiary/tensorflow-jupyter-gpu:latest