The Research Of GPU Application Technology

As a key technology of VR(virtual reality), real-time rendering, is always a hot spot in VR researching. In application such as 3D game, virtual training system, 3D geometry information system, VR researchers are aiming at how to give the user more real experience. The improvement of GPU's computational ability has provided a good hardware foundation. From 1993, every half year, there will be a new generation of GPU developed. The most advanced GPU of now already obtained the similar computational ability of the most advanced CPU of now, even more powerful. The revolution of GPU is more than computational improvement, the programmable pipeline substituting the fixed function pipeline is the most important revolution. It gives users more freedom when doing rendering, also gives GPU some feature of general processor, which gives a much more bigger space of GPU application. With such a fast growing GPU, some real time rendering effect that could not be imaged not a very long time ago has become possible now. It's now a hot spot in VR graphic area about how to make full use of such a computational ability to implement some complex graphic application.My works will be focused on the following topic:First, we have some research on algorithm of real time shadow rendering. We analyzed some current popular shadow algorithm, including the goodness and weakness, and in order to solve the bias problem about shadow maps algorithm, we present a novelty Adaptive Depth Bias Algorithm, which can set the depth bias according the 3D scenes easily. Compared with the traditional algorithms, our algorithm can analyses the characteristic of the 3D scenes and set a suitable depth bias.We also implemented a GPU based particle system. Particle system have been used widely in rendering shape-unfixed objects such as rain, cloud, wind. But the ordinary particle systems can only handle up to 10000 articles in real time simulation, the bottleneck are the transfer of particle data from the CPU to the GPU for rendering and the parallel computational ability of CPU. Using the multiple passage of GPU, it is possible to increase a numerous of particles in the real time simulation to enhance the reality of the scenes. The experiments prove that the particle system based on GPU has a great ability in real time simulation than the ordinary particle systems.The third part of this paper make some research on general computing implementation on GPU. Some scientific computing such as image filtering, requires a powerful parallel computational ability, while the some features of the GPU's ability can meet such a request. We give the GPU implementation of two basic approaches to image filtering– Fast Fourier Transform (frequency domain) and convolution (spatial domain), and evaluate these methods in terms of the performance and suitability for GPU implementation.