CUDA Programming: CPU vs GPU Performance

This mini-project demonstrates the performance difference between CPU and GPU execution by computing a nonlinear mathematical function over 10 million floating-point values.

The function evaluated is:

[ y = sin(x) + log(x + 1) + sqrt{x} ]

Features

• CPU implementation in C++
• GPU implementation in CUDA
• Performance benchmarking and timing comparison
• Run and tested on Google Colab with NVIDIA Tesla T4 GPU

Repository Contents

File	Description
cpu_nonlinear.cpp	C++ source code for CPU implementation
gpu_nonlinear.cu	CUDA source code for GPU implementation
HPC_CUDA.ipynb	Colab notebook containing both implementations and results
AHP_CUDA_Report.pdf	Detailed report including setup, results, and inference

Tech Stack

• C++
• CUDA Toolkit 11.8
• NVIDIA Tesla T4 GPU (Google Colab)

How to Run

You need to have CUDA Toolkit installed (or use Google Colab with GPU enabled).

Compile and Run CPU version

Open a terminal in the project directory and run:

g++ cpu_nonlinear.cpp -o cpu_exec
./cpu_exec

Compile and Run GPU version

Open a terminal in the project directory and run:

nvcc gpu_nonlinear.cu -o gpu_exec
./gpu_exec

Results

Metric	CPU	GPU
Time (seconds)	0.3584	0.0506

Configuration:

• Threads per block: 256
• Blocks per grid: 39,063
• Total GPU threads launched: 10,000,128

Conclusion:
GPU execution achieved approximately 7× speedup over CPU.