Super-Resolution Processing Across GPUs and NPUs for High-Res Displays
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image processing systems struggle to efficiently manage resolution changes, particularly when rendering images on display devices with higher capabilities than initially intended, leading to excessive load on primary GPUs and limitations in achieving enhanced display resolutions.
Innovation Solution
A system that distributes image processing tasks across multiple GPUs and neural processing units (NPUs), offloading super-resolution processing to secondary GPUs or NPUs with trained machine learning models, allowing images to be rendered at desired display resolutions by generating them at lower resolutions on primary GPUs and upscaling them using super-resolution models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If images are rendered directly at high display resolutions using primary GPU, then display quality is improved, but processing load and energy consumption increase excessively
Solution Approach 1:
The image processing task is segmented into two distinct stages: (1) primary GPU generates images at lower resolution, and (2) secondary NPU performs super-resolution upscaling. This division separates the computationally intensive rendering task from the enhancement task, allowing each processor to operate at optimal efficiency levels.
Solution Approach 2:
A lower-resolution intermediate image serves as the mediator between the primary GPU and secondary NPU. The primary GPU produces this intermediate representation, which then becomes the input for the super-resolution model on the NPU, enabling efficient two-stage processing.
2Measurement precision
If images are upscaled using super-resolution models, then display resolution is enhanced, but processing time increases due to additional processing steps
Solution Approach 1:
Traditional geometric upscaling methods are replaced with a machine learning-based super-resolution model on the NPU. This substitution enables higher quality resolution enhancement with better computational efficiency, as the NPU is specifically designed for ML workloads.
Solution Approach 2:
The system dynamically adjusts the input resolution parameter for the super-resolution model based on the target display resolution. By optimizing this parameter, the system achieves the desired output quality while minimizing processing time.
3Measurement precision
If primary GPU generates images at higher resolutions, then final output quality improves, but device complexity and resource requirements increase
Solution Approach 1:
The system employs a multi-functional processing architecture where the primary GPU handles general-purpose rendering at standard resolutions, while the secondary NPU handles specialized super-resolution tasks. This universal design allows the system to adapt to different display requirements without requiring the primary GPU to handle all high-resolution scenarios.
Solution Approach 2:
Instead of requiring the primary GPU to directly produce high-resolution images, the system creates a lower-resolution copy that is then enhanced by the NPU. This copying approach reduces the computational burden on the primary GPU while achieving the same final output quality.
4Adaptability or versatility
If conventional single-GPU processing is used, then system simplicity is maintained, but adaptability to high-resolution displays is limited
Solution Approach 1:
The system adds a new processing dimension by introducing a second computational unit (NPU) dedicated to super-resolution. This dimensional expansion transforms the single-GPU architecture into a multi-processor system that can handle high-resolution displays effectively.
Solution Approach 2:
The processing architecture is designed to be dynamic, automatically selecting between different processing modes based on display requirements. When high-resolution output is needed, the system activates the NPU super-resolution path; otherwise, it uses standard GPU rendering, providing adaptability without constant complexity.
Data Source
AI summary
Systems are configured for performing super-resolution processing for applications by a plurality of different hardware processing units that include hardware processing units that are native to the applications and hardware processing units that are non-native to the applications. Interfaces in the system generate different sets of instructions that are submitted to the different processing units with synchronization objects that synchronize the execution of the instructions.