Performance

Optimization strategies and profiling for N-Body Simulation.

Performance Guide

Optimization strategies and profiling for N-Body Simulation.

Performance Metrics

Benchmarks (RTX 3080)

Particles Direct N² Barnes-Hut Spatial Hash
10K 60+ FPS 120+ FPS 120+ FPS
100K ~10 FPS 60+ FPS 90+ FPS
1M <1 FPS 25+ FPS 60+ FPS
5M N/A ~5 FPS 15+ FPS

Memory Footprint

Configuration Memory Usage
Base (position, velocity, mass) ~52 bytes/particle
Barnes-Hut tree +~32 bytes/particle
Spatial hash grid +~16 bytes/particle
1M particles total ~52-84 MB

Optimization Strategies

1. Algorithm Selection

Choose the right algorithm for your particle count:

1
2
3
4
5
6
7

if (n < 10000) {
    method = ForceMethod::DIRECT;
} else if (long_range_forces) {
    method = ForceMethod::BARNES_HUT;
} else {
    method = ForceMethod::SPATIAL_HASH;
}

2. Time Step Tuning

  • Smaller dt = more accurate, slower
  • Larger dt = faster, potential instability
  • Typical range: 0.0001 to 0.01

3. Softening Parameter

Prevents numerical singularities:

1
2
3

// Too small: numerical instability
// Too large: distorts physics at small scales
config.softening = 0.01f;  // Good default

4. Barnes-Hut Theta

1
2
3
4

// Lower θ = more accurate, slower
config.theta = 0.3f;  // High accuracy
config.theta = 0.5f;  // Balanced
config.theta = 0.7f;  // Fast, lower accuracy

5. GPU Architecture

Compile with your GPU’s compute capability:

1
2
3
4
5
6
7
8

# RTX 30xx (Ampere)
cmake .. -DCMAKE_CUDA_ARCHITECTURES=86

# RTX 20xx (Turing)
cmake .. -DCMAKE_CUDA_ARCHITECTURES=75

# GTX 10xx (Pascal)
cmake .. -DCMAKE_CUDA_ARCHITECTURES=61

Profiling

Nsight Compute

1

ncu --set full ./nbody_sim 100000

Nsight Systems

1
2

nsys profile -o report ./nbody_sim 100000
nsys ui report.nsys-rep

Built-in Timing

Enable in config for frame time output:

1

config.enable_profiling = true;

Best Practices

  1. Always build Release mode: cmake -DCMAKE_BUILD_TYPE=Release
  2. Use appropriate algorithm: See selection guide above
  3. Monitor VRAM: Use nvidia-smi to check memory usage
  4. Batch multiple simulations: Amortize initialization cost
  5. Profile before optimizing: Don’t guess, measure