Memory Architecture

Abstract-As the performance-price ratio of a GPU becoming high, lots of systems may have more than one GPU in a node. Every GPU in a node has the strong ability to render, and it is very important to effectively organize the parallel rendering pipeline to fully exploit the compute units of the system. However lots of parallel rendering systems couple the hardware rendering stage with the composition stage in the display thread and this will make GPUs stall. In this paper, we describe a parallel rendering approach which enables the two stages to execute in parallel. With the frame buffer in the main memory, the full image rendering time totally lies on the GPU rendering ability when the rendering task is large enough. The experiments show that our method performance are much better than that of the existing methods. And we also test the scalable ability and get a linear performance speedup with the GPU number when the rendering task is large enough. Any parallel rendering application can benefit from the decoupled parallel rendering approach