ABSTRACT
We propose a technique for selectively reducing the pixel shader precision for the purpose of efficient rendering in a low-power environment. Typically, pixel shading consumes the largest percentage of the power on GPUs [Pool 2012]. Modern APIs like Direct3D 11.1 allow users to reduce the shading precision to meet the low power requirements, but they don’t allow doing so adaptively. Rendering at reduced precision can potentially produce artifacts in the rendered image. These artifacts are most noticeable where the user’s attention is focused, which is typically at the center of the screen or, in the case of an eye tracking device, at the point provided as input by such a device. This chapter presents a scheme to render the scene at the highest precision where user’s attention is focused and gradually reduce the precision at the points farther away from the focal point.
