PERFORMANCE ANALYSIS OF TRAVERSAL TWO-LEVEL BVH TREES FOR GPU RAY TRACING
L.M. SMIRNOV, Vladimir Alexandrovich Frolov, Alexey Gennadievich Voloboy · 2024
This paper presents a comparative analysis of the performance of several BVH traversal methods on PC(desktop) and mobile GPUs. We examined BVH with different depths and the number of child nodes, implemented several stack traversal algorithms and two different stackless traversal algorithms. We proposed our own stackless traversal option, which is more productive than existing ones in some cases, and our own version of BVH compression of a tree with 2 nodes, which loses no more than 15% of performance. As a result of the performance analysis, we have identified some common problems that occur in almost all algorithms, when they are implemented on GPUs. We believe that our analysis will help developers of ray tracing hardware accelerators to create a more economical hardware solution that is not limited to ray tracing alone. More specifically, the results of our experiments indicate that it is possible to get acceleration up to 5 times by changing the mechanism of the L2 cache, and on desktop GPUs with hardware acceleration of ray tracing, this seems to have already been done not only within the framework of the hardware acceleration of ray tracing itself, but also more generally the case.