Comparison of pseudorandom number generators for use with Monte Carlo-based rendering

Abstract

Path tracing is one of unbiased rendering methods. Said method relies on integral rendering equation to simulate wide variety of optical phenomena. In path tracing, integral part of rendering equation is calculated using Monte Carlo method, which makes the method very resource-intensive. Taking in account that the rendering method is perfectly parallel, it is appropriate to take advantage of GPGPU methods to leverage GPU processing power for the calculations. However, Monte Carlo method requires random numbers to operate, and truly random numbers are difficult to obtain in bulk, especially on GPU. This leads to the crux of the problem: which pseudorandom number generator algorithm would be preferable to use on GPU with Monte Carlo method?

This paper focuses on comparison between several PRNGs (pseudorandom number generators) being used with Monte Carlo method to solve integral rendering equation on GPU. The comparison is used to show advantages and disadvantages to using these PRNGs with Monte Carlo-based rendering methods, and draws some conclusions regarding picking correct tool for the job.