Traversal methods for GPU ray tracing

Ways of exploiting the raw performance of GPUs for computing ray tracing have been a hot research topic recently. Performance similar to the multi-core CPU ray tracing engines has been achieved. In this paper we present a new traversal method created by combining two of the existing methods. The proposed method is less sensitive to performance loss due to certain object scene distribution. It can also be faster than the methods from which it originated. Several possibilities how to create such a method exist and even better methods can be constructed with the addition of a single instruction to the next generation of GPUs. The resulting ray tracer achieves 50+ fps for primary rays on moderately complex scenes running on current mainstream GPUs.