Performance comparison of parallel turbulent noise evaluation with different gradient selection methods

Sammanfattning: Noise is of vital interest in many parts of computer sciene, especially in the computer graphics eld where noise is used to create nature-like e ects. Perlin’s 1985 algorithm to generate noise remains the most pop- ular in spite of many alternatives having been presented over the years. In this report we have examined the execution time impact of two new gradient table data structures and a new hash method for this algo- rithm, suggested by Perlin in 2002 and Olano in 2005 respectively. Our implementation simulated turbulence and ran in parallel on a modern GPU using the OpenCL framework. We also examined if the turbulence method’s octave summation could bene t from parallelization. Results suggest that Olano’s hash method performs signi cantly faster, while Perlin’s original gradient table data structure performs slightly faster than the suggested improvements. We also found that a paral- lelization of the octave summation in the turbulence method performs signi cantly faster.