Biography
Enrollment Date: 2012
Graduation Date：2015
Degree：M.S.
Defense Date：2015.06.02
Advisors：Yangdong Deng
Department：Institute of Microelectronics，Tsinghua University
Title of Dissertation/Thesis：Ray Tracing Rendering Engines
Abstract:
Ray tracing is considered to be a promising rendering technology, which can produce physically realistic pictures. It is widely used by advanced graphics applications, such as movie special-effects and the computer aided design. Traditional ray tracing algorithms are usually based on high performance computing platforms, while the feasibility of mobile GPU-based ray tracers is investigated for the first time in this paper.This paper presents a set of ray tracing algorithms, and implements them on IMAGINATION PowerVR SGX 544-MP3 GPU and NVDIA Tegra K1 GPU, using OpenCL and CUDA. It then analyzes the program performance bottlenecks and tests the input power of the system. Next, a detailed characterization is performed, involving the achieved occupancy, the instructions efficiency, as well as the memory bandwidth of the mobile GPUs. It also investigates the impact of memory usage on the system’s performance and power consumption. In the end, all these results are compared against the corresponding implementations on ARM CPU and desktop GPU, and optimization strategies are presented. It is proved that the Tegra K1 GPU already allows constructing the acceleration structure of an 1M-triangle scene within 120ms. At the same time, compared to the serial ray traversal program, which runs on a 2.3GHz ARM Cortex A15 CPU, using the algorithm frameworks presented in this paper can speed up our mobile ray tracer 100 times over, with a throughput of 7 million rays per second.