Dynamic rendering quality scaling for mobile GPU

Abstract

As display resolution exponentially increases in mobile platforms, maintaining acceptable frame rate (e.g., 60Hz) becomes more challenging than ever because elevated computing demand (for both CPU and GPU) results in reduction of battery use time and increase in device surface temperature. Because power consumption increases almost linearly with GPU workload, the heavy GPU workload imposed by fixed high resolution can be allowed to be actually computed only when there are human perceptible benefits. Some recent techniques are exploited at runtime to reduce the workload such as skipping frames or lowering resolution during rendering. From those techniques, we still observe noticeable artifacts because they did not carefully investigate whether resulting sequences of frames are perceived as containing artifacts or not. Although one of recent studies relies on a user's view distance [Nixon et al. 2014], it can malfunction depending on environmental and biometric limitation.