Variable Resolution Rendering for VR Graphics Workload Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current graphics processing units (GPUs) face challenges in rendering high-quality images in real-time, especially in virtual reality (VR) and mobile devices, due to limited compute power and bandwidth, which is exacerbated by the need for high resolution and refresh rates, while human visual limitations suggest that not all areas require high resolution for optimal user experience.
Innovation Solution
The technique modifies Tile-Based Rendering by calculating a down-scale factor for each tile based on its prominence, using a lens distortion map, fovea location, and color variation to render at varying resolutions, minimizing hardware workload and storing intermediate images in on-chip memory for efficient upscaling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high resolution rendering is used for all areas, then image quality is improved, but GPU workload and power consumption increase significantly
Solution Approach 1:
The patent applies local quality by rendering different regions of the image at different resolutions based on their visual importance. The image is divided into tiles, and each tile is rendered at a resolution appropriate to its prominence, allowing high quality where needed while reducing workload in less critical areas.
Solution Approach 2:
The patent segments the image into multiple tiles and processes each tile independently with its own resolution setting. This segmentation allows the system to apply different rendering qualities to different regions, optimizing the balance between image quality and computational load.
2Measurement precision
If high resolution rendering is used for all areas, then image quality is improved, but rendering speed decreases due to increased GPU workload
Solution Approach 1:
The patent renders each tile at a resolution matched to its visual importance rather than uniformly high resolution. This local quality approach maintains image quality in prominent areas while significantly reducing the total number of pixels rendered, thereby improving rendering speed.
Solution Approach 2:
By segmenting the image into tiles and rendering them at different resolutions, the patent reduces the total computational workload while maintaining overall image quality. This segmentation strategy enables faster rendering by avoiding unnecessary high-resolution processing in less important regions.
3Productivity
If variable resolution rendering is implemented, then GPU workload is reduced, but device complexity increases
Solution Approach 1:
The patent implements dynamic resolution adjustment by calculating a scale factor for each tile based on its prominence. This dynamic approach allows the rendering system to adaptively adjust resolution per tile, reducing GPU workload while managing complexity through algorithmic rather than hardware solutions.
Solution Approach 2:
The patent changes the resolution parameter dynamically for each tile based on calculated prominence metrics. By modifying the rendering resolution parameter on a per-tile basis rather than using fixed hardware configurations, the system reduces GPU workload while avoiding significant device complexity increases.
Data Source
Figure 1~2
Figure 3~4
Figure 5~5(b)
AI summary
A device for rendering an image in a graphics processing system, the image being divided into a plurality of tiles, each tile corresponding to one or more display pixels and at least some of the tiles containing at least one primitive of an object, the device being configured to: calculate, for a tile, in dependence on the prominence of the tile, a scale factor for the respective tile; and, for at least some of the tiles: shrink the primitive(s) of the tile to a sub-area of the tile, the site of the sub-area being dependent on the calculated scale factor for the tile; render the whole tile to form an intermediate image; generate display pixels by upscaling the intermediate image.