Innovative GPU based Matrix Optimization for Sustainable Real Time Rendering in AR/VR Systems

Abstract

Real time rendering is a foundation of more immersive systems such as augmented reality and virtual reality, in which high performance capabilities in the form of high rates and low latency are required. In AR/VR systems, geometrical transformation, camera modelling, skeletal animation, and projections entail high computational loads in the form of intensive calculations involving matrices. As a result of increased complexity, limitations, and device constraints, high performance and energy efficiency levels required in corresponding calculations can be a serious hindrance. In this paper, new approaches in optimizing AR/VR related GPU hardware will be discussed and introduced. Utilizing parallel processing architectures and computing through memory hierarchy in GPUs will allow new approaches in developing efficient approximations in calculations requiring heavy matrix processing. The paper presents a theoretical model to transform, multiply through matrices, and approximate in calculations. The paper establishes a very strong connection between concepts of linear algebra and high-performance needs in AR/VR rendering techniques. The experimental results in AR/VR systems show new approaches in designing optimized systems with capabilities in energy efficient computing.