Performance Analysis of Parallel Matrix Computation (CPU vs GPU)

📌 Project Overview

This project demonstrates the practical benefits of High Performance Computing (HPC) by benchmarking large-scale matrix multiplication on CPU vs GPU.

A comparative study is performed between:

• CPU-based serial computation using NumPy
• GPU-accelerated parallel computation using CuPy (CUDA)

The project is designed to run out-of-the-box on Google Colab using an NVIDIA Tesla T4 GPU.

🚀 Motivation

In modern scientific computing, machine learning, and data analytics, sequential CPU execution often becomes a performance bottleneck. This project aims to:

1. Quantify the performance improvement achieved through GPU acceleration
2. Demonstrate CUDA-enabled parallel computing using high-level Python libraries
3. Provide an academically sound, reproducible reference for MSc-level HPC coursework

🛠 Technologies Used

• Language: Python 3.10
• CPU Computation: NumPy
• GPU Computation: CuPy
• Parallel Platform: NVIDIA CUDA
• Execution Environment: Google Colab (Jupyter Notebook)

📂 Repository Structure

HPC-Matrix-Benchmark-GPU/
├── colab_notebook.ipynb        # MAIN FILE (Run on Google Colab)
├── README.md                   # Project Documentation
├── requirements.txt            # Optional local dependencies
├── src/
│   ├── cpu_version.py          # CPU implementation (NumPy)
│   ├── gpu_version.py          # GPU implementation (CuPy)
│   └── benchmark.py            # Benchmark driver
├── results/                    # Benchmark logs / outputs
└── report/
    └── HPC_Project_Report.md   # Full academic project report

⚡ How to Run

Option 1: Google Colab (Recommended)

1. Download colab_notebook.ipynb from this repository
2. Upload it to https://colab.research.google.com
3. Go to Runtime → Change runtime type
4. Select GPU (T4) as the hardware accelerator
5. Click Runtime → Run all

✅ No local setup required

Option 2: Local Execution (Requires NVIDIA GPU)

Requires:

• NVIDIA GPU
• CUDA drivers
• Compatible CuPy installation

# Clone the repository
git clone https://github.com/partha392/HPC-Matrix-Benchmark-GPU.git
cd HPC-Matrix-Benchmark-GPU

# Install dependencies
pip install -r requirements.txt

# Run benchmark
python src/benchmark.py

📊 Sample Output

(Observed on Google Colab with NVIDIA Tesla T4 GPU)

Matrix Size: 4000 x 4000
CPU Time: 12.8 – 18.1 sec
GPU Time: 0.08 – 0.15 sec
-------------------------
Observed Speedup: ~80x – 180x

⚠️ Exact performance may vary due to shared cloud infrastructure and runtime conditions.

📜 Academic Report

A detailed academic report is available in the report/ directory, covering:

• HPC and CUDA architecture overview
• Experimental methodology
• Performance benchmarking and speedup analysis
• Limitations and future scope

📄 File: report/HPC_Project_Report.md

🤝 Contributions & Extensions

This is an academic HPC benchmark project. Possible extensions include:

• Multi-GPU benchmarking (NCCL)
• Mixed-precision computation using Tensor Cores
• Additional benchmarks (FFT, Monte Carlo, reductions)

Suggestions and improvements are welcome.

📝 License

This project is licensed under the MIT License and is free to use for educational and academic purposes.