GPU Benchmarking

This project provides a comprehensive benchmarking and analysis framework for GPU operations. It includes implementations using naive methods, NumPy, and PyTorch, and offers detailed performance measurement and visualization through a PyQt5-based dashboard.

Features

• Benchmarking GPU-related operations with multiple implementations:
  • Naive (manual) implementation
  • NumPy-based optimized operations
  • PyTorch-based GPU acceleration
• Supports key GPU operations including:
  • Matrix multiplication (matmul)
  • Matrix inversion (matinv)
  • Distance calculations (dist)
• Performance measurement using execution time and memory usage
• Detailed analysis and visualization via a PyQt5 GUI dashboard
• Data management and result export functionality
• Support for comparing multiple methods across different matrix sizes

Project Structure

.  
│  
├── main.py              # Main GUI application  
├── proj.py              # Core benchmarking functions  
├── gpu_monitor.py       # GPU monitoring utility  
├── analyse.py           # Data analysis and visualization  
├── results.csv          # Benchmark results  
└── gpu_log.csv          # GPU monitoring logs

Prerequisites

• Python 3.10 or higher

• Required Python packages (install via pip):

  pip install numpy pandas matplotlib seaborn pyqt5 pytorch

How to run

Run Dashboard.py to perform benchmarking, storing data, and visualasing data and analysing:

python Dashboard.py

Results are saved in results.csv and can be reloaded for later analysis.

Contributing

If you find any bugs or have ideas to make this project better, just send a pull request.

License

This project is licensed under the MIT License.

Authors

AmirMasoud Ebrahimi, Alireza Mohandesi, Sajjad Agheli, AmirReza Jafari