Semi-Static Load Balancing for Low-Latency Ray Tracing on Heterogeneous Multiple GPUs

Ray tracing is used to render a realistic image but the drawback is its high computational cost. Although there are studies accelerating ray tracing using the GPU, even with the latest GPU, we cannot get a satisfactory rendering speed. An obvious way to accelerate it further is to use more than one GPU. To exploit the computational power of multiple GPUs, the work has to be distributed in a way so that it minimizes the idle time of GPUs. There are studies on load balancing CPUs, but they are not directly applicable to multiple GPUs because of the difference of the architectures, as discussed in Section 22.2.