| Summary: | 3D computer graphics hardware has assumed increased significance with the emergence of high-quality and interactive 3D graphics in many applications. Due to intensive floating point and fixed point-computations and high memory bandwidth requirements, 3D applications are already one of the major power consumers of today's computing devices, such as PCs, workstations, and games. As the demand for higher quality-rendering grows, this problem becomes more acute with the migration of graphics into power-constrained notebook computer and post-PC systems, such as wearable computer, personnel digital assistant(PDA), web-pad, and 3D graphics heads-up displays. This dissertation explores approximate signal processing (ASP) based architectures and systems, and proper management of ASP systems based on the relationship between energy and perceptual quality (E-Qp) in order to reduce the power consumption of real-time 3D graphics rendering systems. Our works are based on the observation of the E-Qp relationship which varies along with the variation of the contents to be perceived by human visual system. One of the human visual perception (HVP) characteristics is that visual sensitivity varies according to spatial and temporal frequency. By applying lower quality algorithms or less accurate computations when HVP is less than a predetermined threshold, it is possible to keep perceivable quality constant while consuming less power. The novelty of this ASP system is that it can be dynamically configured to operate at various degrees of quality depending not only on the system resources like a conventional ASP system, but also on the E-Qp relationship. This dynamic configuration is achieved by using algorithm level configuration and parameter configuration in terms of reconfiguration pattern. This dissertation demonstrates the effectiveness of the above proposed reconfigurable ASP system in obtaining power savings from the shading and texture mapping blocks which are the main function blocks of the 3D graphics rendering system. More specifically, algorithm level configuration is used for adaptive shading and texture mapping, and parameter level configuration is used for 3D vector interpolator of Phong shading algorithm. Power savings is measured using a hardware synthesis and estimation methodology with realistic graphics content from the well-known 3D graphics benchmarks and the game Quake. Power savings of up to 72%, 75.1%, and 73.8% are obtained from the CORDIC normal vector interpolator, the adaptive shading, and the adaptive texture mapping, respectively.
|
|---|
