Scalable Parallel Path Tracing
2019, Ikkala, J., Bachelor of Science thesis, Tampere University.
Abstract: Parallelized path tracing for real-time, low-latency content over a network connection is a potential future direction for mobile gaming and other 3D graphics consumption. We have implemented a 3D renderer that parallelizes path tracing to multiple computation devices, including remote servers in addition to local resources. This thesis describes a 3D renderer that is able to dynamically adjust to uneven and changing hardware performance and can use both local resources and server resources for path tracing and noise removal. The renderer is able to keep input-to-image latency below 50 ms, gain significant benefit from server resources and keep image quality acceptable. Quality losses from compromises made for parallelization are around 1 dB to 2.5 dB. These values point to a possibility of streaming high-quality path traced content over a sufficiently fast and high bandwidth network, such as the upcoming 5G mobile network standard. Virtual reality-compatible latency can be achieved by situating computation capacity close to the user to minimize physical limitations on delay.