Understanding the efficiency of ray traversal on GPUs