GPU Programming: CUDA & HIP (NVIDIA & AMD)

This repository contains GPU-based parallel programming examples using CUDA and HIP frameworks. The goal is to compare and demonstrate the behavior of GPU programs across different architectures—NVIDIA and AMD—using CUDA and HIP in both CUDA and ROCm environments.

🚀 Frameworks Used

• CUDA (Compute Unified Device Architecture) – for NVIDIA GPUs
• HIP (Heterogeneous-Compute Interface for Portability)
  • HIP on NVIDIA GPU (compiled with hipcc using CUDA backend)
  • HIP on AMD GPU (compiled with hipcc using ROCm backend)

⚙️ Environment Details

• CUDA Version: 12.5
• HIP Version: ROCm 6.x (for AMD), HIP with CUDA backend (for NVIDIA)
• GPUs Used:
  • NVIDIA T1000 8GB
  • AMD Radeon (ROCm Supported) MI50/MI60
• Development Platform:
  • Google Colab (CUDA)
  • ROCm Environment on AMD Server
  • AlmaLinux 9.5 (Host System)

📖 How to Run

Make sure to have the appropriate GPU and drivers installed (CUDA or ROCm). [The compilation command may vary on the basis of the system being used and CUDA Version, therefore, check it before using]

To compile and run the CUDA Programs

To compile the CUDA Programs- nvcc -Wno-deprecated-gpu-targets filename.cu -o outputfile

To run the CUDA Programs- ./outputfile

To compile and run the HIP Programs

For NVIDIA (CUDA backend)

To compile- hipcc filename.cpp -o outputfile

To run- ./outputfile

For AMD (ROCm backend)

To compile- hipcc filename.cpp -o outputfile

To run- ./outputfile

Purpose

This project is developed as part of a Parallel Programming Comparative Study to understand GPU computing frameworks and their behavior across different architectures.

Feel free to raise an issue or reach out if you have any questions or suggestions!