Research Of The Parallel Graphics Rendering Technology Oriented To High Performance Supercomputers

With the improvement of theories and technologies of compute graphics, the technology of graphics rendering plays an important role in many society fields such as design, entertainment, science and etc. As the same time, high-performance computing applications of nuclear, meteorology, aircraft simulation and high energy physics simulation in the progressive society keep challenging the technologies of graphics rendering for the more complicated scene dataset on many aspects: higher rendering speed, higher resolution and more realistic. Using the high performance supercomputer to do the graphics calculation processing of the large-scale, time-varying, high-dimensional, complicated datasets is an important way to implement the high performance graphics rendering and the high resolution display.This dissertation has a research on the technology of the parallel graphics rendering oriented to high performance supercomputers. The main work is listed as following.1. A parallel graphics rendering system oriented to high performance supercomputers is designed and implemented. Its hardware platform is built with the high performance supercomputer and the graphics rendering service cluster, on which an immediate-mode and sort-first parallel graphics rendering system is developed. The system has speeded up the frame rendering rate by making use of the former to share the computing overhead of the latter.2. A parallel rendering stream processing mechanism oriented to high performance supercomputers is presented. It introduces the concept of the stream processing which uses limited memory space to deal with the potential unlimited graphics input sets. It has increased the flexibility and adaptability of the parallel rendering system by using some technoloy such as modularization and transparently parallelization.3. A multi-screen tiled display technology for the parallel graphics rendering system oriented to high performance supercomputers is deeply surveyed. Some strategies such as the visibility culling belonging judgement on the geometry data in the scene and the lazy buffer update are presented for the network bottleneck problem, which has reduced the amount of network transmission and the times of network connections to implement the real-time high resolution display.4. A dynamic load balancing method is presented for the parallel graphics rendering system oriented to high performance supercomputers, which adjusts the number of service nodes for each tile according to the load distribution in every frame to balance the load on the service nodes.At last, by using some of the research results mentioned above, a parallel graphics rendering prototype system is implemented and then tested on its correctness and applicability of the large scale complicated scene rendering and the multi-screen tiled diplay.