
In the paper, the limitations of ray tracing performance due to the way NVIDIA GPUs are currently designed are discussed. At present, in order to meet the needs of large-scale parallel computing, the GPU uses the SIMD execution mode. The input of several identical operations will be packaged into a group for parallel execution. This group is the smallest execution unit of the GPU, which Nvidia calls “warp”. The GPU hides the pause by mobilizing the warp, but in real-time ray tracing operations, problems may occur and performance losses may occur. And “GPU Subwarp Interleaving” is a solution to the current GPU dilemma.
Using a suite of applications with ray-tracing workloads, the researchers achieved performance gains of an average of 6.3 percent and up to 20 percent on a modified, enhanced Turing architecture GPU. Existing GeForce graphics cards won’t be able to get this effect with a driver update, only this new technology will be applied to future architectures.