N-Body CUDA Simulation

This project implements an N-Body simulation using CUDA for parallel computation. The simulation calculates the gravitational interactions between a large number of bodies, making use of GPU acceleration to handle the computational load efficiently.

Project Description

The N-Body problem involves predicting the individual motions of a group of celestial objects interacting with each other gravitationally. This project was developed as part of a university course to demonstrate the use of CUDA for high-performance computing.

Prerequisites

Before you begin, ensure you have met the following requirements:

• A CUDA-capable GPU
• NVIDIA CUDA Toolkit installed
• CMake version 3.10 or higher
• A C++ compiler compatible with CUDA

Building the Project

To build the project, follow these steps:

1. Clone the repository:

   git clone <repository-url>
   cd N-Body-CUDA

2. Create a build directory and navigate into it:

   mkdir build
   cd build

3. Run CMake to configure the project:

   cmake ..

4. Build the project using Make:

   make

Usage

Running the Simulation

To run the N-Body simulation, use the following command:

./nbody_sim_cuda <initial_states_file> <output_file> <dt> <num_steps> <block_size>

• <initial_states_file>: Path to the file containing the initial states of the bodies.
• <output_file>: Path to the file where the simulation results will be saved.
• <dt>: Time step for the simulation.
• <num_steps>: Number of simulation steps.
• <block_size>: CUDA block size.

Running the Benchmark

To run the benchmark, use the following command:

./nbody_cuda_benchmark <num_bodies> <dt> <num_steps> <block_size> <method>

• <num_bodies>: Number of bodies in the simulation.
• <dt>: Time step for the simulation.
• <num_steps>: Number of simulation steps.
• <block_size>: CUDA block size.
• <method>: Simulation method (0 for standard, 1 for velocity Verlet, 2 for optimized).

Visualization

The project includes a simple Python-based visualization using VisPy to visualize the trajectories generated by the simulation.

Prerequisites

Before running the visualization, ensure you have the following Python packages installed:

• numpy
• vispy
• seaborn

You can install these packages using pip:

pip install numpy vispy seaborn

Running the Visualization

To run the visualization, use the following command:

python visualization_canvas.py

Make sure the initial_conditions.txt and trajectories.txt files are in the same directory as the visualization_canvas.py script. These files should contain the initial conditions and the trajectories generated by the simulation, respectively.

Implemented Kernels

The project implements three different CUDA kernels for the N-Body simulation:

1. Standard N-Body Simulation: Each thread updates a single body by calculating the gravitational forces from all other bodies.
2. Velocity Verlet Integration: Similar to the standard method, but uses the Velocity Verlet algorithm for better numerical stability.
3. Optimized N-Body Simulation: Each thread calculates each physical interaction between bodies. This method aims to be faster but currently does not work properly and is very slow.

Performance Comparison

The following plot compares the compute time on CPU (Ryzen 9 3900X) and GPU (RTX 4090) with a block size set to 256.

Acknowledgments

This project was developed during university classes. Special thanks to the course instructors and teaching assistants for their guidance and support.