Abstract

During this thesis, we developed a proof-of-concept implementation of a ‘shared rendering computation’ infrastructure. The implementation consists of a server written in Vulkan that computes the global diffuse illumination, and a client web-application that receives this illumination information in order to enhance its real-time rendering capabilities. By doing this, we aim to support devices with low-end rendering hardware to render computationally intensive globally illuminated scenes interactively. Moreover, the otherwise redundant rendering workload is only being processed once for all client instances. The rendering algorithm is based on a virtual light field approach to global illumination. It exploits the use of the fast rasterization hardware available on GPUs to propagate parallel bundles of rays throughout the scene. Visibility is handled by multi-sampled rendering into a depth buffer. The proposed architecture scales well in a densely populated Cloud Gaming environment, compared to linear scaling with traditional client-side rendering.

Resources

Thesis

Demo:

BibTex reference

@misc{vandersanden2021shared,
        author = {Vandersanden, Jente},
        title = {Shared Rendering Computation in Cloud Gaming},
        year = {2021}
}