A scalable and coherent approach to Monte Carlo path tracing for multi-user environments

Abstract

We present an adaptation to Monte Carlo path tracing with the goal of rendering interactive multi-user environments in a scalable manner. The proposed algorithm computes diffuse global illumination in a server environment and caches results for reuse between multiple users. We experimentally evaluate our algorithm on a set of medium to large scenes by comparing its workload to the work performed by a reference GPU path tracer. While contemporary cloud gaming solutions require a separate GPU per user, our algorithm just requires a fixed number of GPUs, determined by the size of the scene. A second topic of focus is the problem of ray incoherence that is natural to light transport simulation. This issue is addressed by adopting a hybrid approach in which nearby lighting is path traced and distant lighting is approximated by radiance probes. Our experiments confirm that a smaller tracing range results in more coherent memory accesses. With the insights gained from our experiments, we propose several optimizations.