Maximizing performance gain of Variable Rate Shading tier 2 while maintaining image quality : Using post processing effects to mask image degradation

Sammanfattning: Background. Performance optimization is of great importance for games as it constrains the possibilities of content or complexity of systems. Modern games support high resolution rendering but higher resolutions require more pixels to be computed and solutions are needed to reduce this workload. Currently used methods include uniformly lowering the shading rates across the whole screen to reduce the amount of pixels needing computation. Variable Rate Shading is a new hardware supported technique with several functionality tiers. Tier 1 is similar to previous methods in that it lowers the shading rate for the whole screen. Tier 2 supports screen space image shading. With tier 2 screen space image shading, various different shading rates can be set across the screen which gives developers the choice of where and when to set specific shading rates. Objectives. The aim of this thesis is to examine how close Variable Rate Shading tier 2 screen space shading can come to the performance gains of Variable Rate Shading tier 1 while trying to maintain an acceptable image quality with the help of commonly used post processing effects. Methods. A lightweight scene is set up and Variable Rate Shading tier 2 methods are set to an acceptable image quality as baseline. Evaluation of performance is done by measuring the times of specific passes required by and affected by Variable Rate Shading. Image quality is measured by capturing sequences of images with no Variable Rate Shading on as reference, then with Variable Rate Shading tier 1 and several methods with tier 2 to be compared with Structural Similarity Index. Results. Highest measured performance gains from tier 2 was 28.0%. The result came from using edge detection to create the shading rate image in 3840x2160 resolution. This translates to 36.7% of the performance gains of tier 1 but with better image quality with SSIM values of 0.960 against tier 1’s 0.802, which corresponds to good and poor image quality respectively. Conclusions. Variable Rate Shading tier 2 shows great potential in increasing performance while maintaining image quality, especially with edge detection. Postprocessing effects are effective at maintaining a good image quality. Performance gains also scale well as they increase with higher resolutions.