A Multi-Engine Benchmark for Relativistic Raytracing

This dataset provides a comprehensive performance and accuracy benchmark for interactive relativistic ray tracing. It consists of 16,656 timed renders, each paired with its corresponding output PNG image. The data evaluates rendering workloads across a factorial design of different architectural and physical configurations.

The dataset captures computational timings and image outputs across the following dimensions:

• Platforms: Implementations built in Unity and Vulkan.

• Spacetime Metrics: Newton (a non-relativistic heuristic baseline), Schwarzschild (static, spherically symmetric black hole), and Kerr (rotating black hole).

• Numerical Integrators: The first-order Euler method tested at three step sizes (260 km, 52 km, and 1 km) and the fourth-order Runge-Kutta (RK4) method tested at a 1 km step size.

• Resolutions: Output dimensions of 144p, 480p, and 720p.

• Scene Complexity: Six distinct scenes that scale in multi-body complexity. These range from single black hole environments with various backgrounds and accretion disks, to binary black hole systems, up to a ten-body solar system.

• Physical Variables: Renders using different gravitational multipliers (Normal, Strong, Max) and Kerr spin parameters (Moderate, High, Very High, Extreme).