Efficient use of Multiple Hardware Components for Image Synthesis

Abstract

With the advent of affordable and powerful computers systems, even in desktop configurations, the rapid render­ing of complex scenes using global illumination algorithms is within reach. However, some significant problems still must be dealt with when performing such physically based lighting simulations i we are to achieve real time performance on current technology. The power offered by modern computer systems is not only due to the in­creasing CPU capabilities, but also to the fast development of others system elements, including graphics proc­essing units (GPU) and other additional processing boards. Ray tracing implementations in such environments typically target only one hardware component, or at best map specific type of work to particular elements within the system. This approach leads to an unbalanced work distribution and marginalisation of certain types of resources. Jn this paper we present a framework for managing and interfacing all the available computational power within a modern PC in a balanced and efficient way. At the centre of proposed approach is an abstraction layer between the main rendering stage and the primitives of the rendering process. Several advantages of such organisation are discussed, including portability and ea.ry extendibility to newer powerful hardware resources as they become available, optimizations at resource level, modularity, and load balancing. More practical issues regarding the management strategy and implementation for heterogeneous environments below this level o ab­ straction, where performance is a key aspect, are also presented. To validate this model a set of experiments were conducted. The results have shown that the introduction of an abstraction layer into the rendering system can improve the performance and better use the available resources.