Porting a Monte Carlo volume rendering method into an open-world AAA game engine

Abstract

Rendering volumetric cloudscapes in real-time is computationally intensive, yet essential for open-world games. The high cost of real-time volumetric cloud rendering primarily stems from the ray marching algorithm, a volume rendering method characterized by a time complexity of $O(n^2)$. An alternative approach in CPU rendering, known as Decoupled Ray Marching [Kulla 2011], optimizes the original algorithm, reducing its complexity to $O(n\log n)$ by decoupling the computation of in-scattering from transmittance and incorporating importance sampling to approximate the final color. The main contribution of this thesis is the adaptation of Kulla's Decoupled Ray Marching algorithm for GPU execution inside Avalanche's Apex Game Engine. This study effectively demonstrates that Kulla’s decoupled ray marching algorithm can be successfully implemented on the GPU, resulting in significant rendering time reductions of up to 30% compared to standard ray marching. However, further research will be necessary to address issues related to noise and optimized GPU memory access.

Resources

• 📄 Scientific paper (PDF)
• 📈 Thesis presentation (PDF)
• 🧑🏻‍💻 Pseudocode (HLSL)

Summary of the Results

Decoupling lighting outperformed nested ray march render time

Especially in partially cloudy conditions, indicating it handles cloud-free areas better.

More memory had a marginal impact on visual quality

The size of the buffers had a smaller impact on rendering time than anticipated, particularly when compared to its effect on visual quality.

Visual quality was sensitive to light samples, but render time was not

The effects on visual quality are significant and noticeable.

How to cite this work

• D. Giraldo Mosquera, “Faster Volumetric Cloudscapes with GPU-based Decoupled Ray Marching : Porting a Monte Carlo volume rendering method into an open-world AAA game engine,” KTH Royal Institute of Technology, Stockholm, 2025. [Online]. Available: https://urn.kb.se/resolve?urn=urn:nbn:se:kth:diva-368116

References

• Christopher Kulla. 2011. Decoupled ray-marching of heterogeneous participating media. In ACM SIGGRAPH 2011 Talks (New York, NY, USA, 2011-08-07) (SIGGRAPH ’11). Association for Computing Machinery, 1. 978-1-4503-0974-5 https: //doi.org/10.1145/2037826.2037900
• Christopher Kulla and Marcos Fajardo. 2012. Importance Sampling Techniques for Path Tracing in Participating Media. 1519–1528. https://doi.org/