Hybrid Rendering Tile Classification for GPU Memory Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current rendering technologies face inefficiencies due to excessive access to external GPU memory, leading to increased power consumption and memory bandwidth issues, particularly when combining rasterization and ray tracing techniques.
Innovation Solution
A hybrid rendering method that subdivides a 2D frame into virtual tiles, classifies them based on reflective or transparent properties, and uses ray tracing for tiles with such properties while employing rasterization for others, storing geometric information in GPU internal memory to reduce external memory access.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If ray tracing rendering is used to simulate physical properties of light for high quality image generation, then image quality is improved, but power consumption increases due to excessive access to external GPU memory
Solution Approach 1:
The patent divides the frame into multiple tiles and further segments them into first tiles (requiring ray tracing for reflective/transparent objects) and second tiles (suitable for rasterization). This segmentation allows selective application of rendering techniques, reducing overall power consumption while maintaining image quality where needed.
Solution Approach 2:
The patent applies different rendering qualities to different regions of the frame. Ray tracing with high computational quality is applied only to first tiles containing reflective or transparent objects, while second tiles use standard rasterization. This local quality approach maintains overall image quality while reducing power consumption.
2Measurement precision
If ray tracing rendering is used to generate high quality image by tracing light paths, then image quality is improved, but memory bandwidth problems occur due to excessive access to external GPU memory
Solution Approach 1:
The patent segments the frame into tiles and classifies them into first and second tiles. By limiting ray tracing to only first tiles (those containing reflective or transparent objects), the patent reduces the total volume of data that needs to be accessed from external GPU memory, thereby alleviating memory bandwidth problems.
Solution Approach 2:
The patent applies ray tracing only partially - specifically to first tiles containing reflective or transparent objects - rather than to the entire frame. This partial application of the computationally intensive ray tracing technique reduces memory bandwidth consumption while maintaining image quality for the critical portions of the scene.
3Productivity
If the frame is subdivided into tiles and classified based on primitive properties, then rendering efficiency is improved by reducing external memory access, but device complexity increases
Solution Approach 1:
The patent introduces a tile subdivision and classification mechanism that divides the frame into manageable tiles and categorizes them based on the presence of reflective or transparent objects. This segmentation enables more efficient rendering by reducing external memory access, and the added processing complexity is localized to the classification stage rather than affecting the entire rendering pipeline.
Data Source
AI summary
A hybrid rendering method includes subdividing a two-dimensional (2D) frame into virtual tiles (hereinafter, referred to as ‘tiles’); classifying the tiles into first tiles which include an object having reflective or transparent properties and second tiles which do not include the object having reflective or transparent properties; storing geometric information of a primitive in the first tiles in a graphics processing unit (GPU) internal memory; and generating a first image by rendering one or more of the first tiles via ray tracing using the geometric information, generating a second image by rendering one or more of the second tiles via rasterization, and outputting a final image by merging the first and second images.