Neural MFSR Precision Switching for Stable Gaming Frame Rates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing multi-frame super resolution (MFSR) techniques in video gaming suffer from high computational demands, leading to reduced frame rates and inconsistent resolution, especially in fast-paced games, causing a less stable and responsive gaming experience.
Innovation Solution
A method and system that dynamically adjust the precision of weights and activations in a trained neural network based on processing unit usage, using multiple neural networks with varying precisions to maintain a stable frame rate without changing resolution, by quantizing weights and activations when high computational load is detected.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multi-frame super resolution is used to increase graphics resolution, then image quality is improved, but computational time increases and frame rate decreases
Solution Approach 1:
The patent applies dynamics by making the neural network precision adaptive rather than fixed. The system dynamically switches between different precision levels (e.g., FP16, FP32, INT8) of the neural network weights and activations based on real-time computational load monitoring, allowing the resolution enhancement process to adjust its computational intensity flexibly during gameplay.
Solution Approach 2:
The patent changes the precision parameter of the neural network weights and activations to resolve the contradiction. By varying the precision level as a controllable parameter, the system can reduce computational demands during high-load scenarios while maintaining acceptable image quality, thus improving frame rate without permanently sacrificing resolution enhancement effectiveness.
2Productivity
If dynamic resolution is used to maintain frame rate during high load points, then frame rate stability is improved, but resolution consistency deteriorates and input lag increases
Solution Approach 1:
Instead of changing the resolution parameter, the patent changes the precision parameter of the neural network. This allows the system to maintain consistent resolution output while adjusting computational precision to manage frame rate during high-load points, avoiding the resolution fluctuation and input lag issues associated with traditional dynamic resolution methods.
Solution Approach 2:
The patent creates multiple versions (copies) of the neural network with different precision levels. These precision variants serve as substitutes for each other based on computational load, allowing the system to maintain resolution consistency while managing performance through precision switching rather than resolution switching.
3Measurement precision
If high precision neural network weights and activations are used for MFSR, then image quality is improved, but computational load increases
Solution Approach 1:
The patent makes the neural network precision dynamic by monitoring computational load in real-time and adjusting precision levels accordingly. During low-load periods, high precision (FP32) is used for maximum image quality. During high-load periods, the system automatically switches to lower precision (FP16, INT8) to reduce computational demand while maintaining acceptable quality levels.
Solution Approach 2:
The patent changes the precision parameter of neural network weights and activations to balance image quality and computational load. By treating precision as a variable parameter rather than a fixed setting, the system can optimize the trade-off between inference accuracy and computational resources based on real-time gaming conditions.
Data Source
AI summary
A computer-implemented method of generating a multi-frame super resolution, MFSR, graphics output for a video gaming system during gameplay, comprising: using a trained artificial neural network, ANN, comprising a plurality of weights and activations to perform multi-frame super resolution, MFSR, based on input graphics data from a game deployed on a video gaming system, to generate a MFSR graphics output; monitoring usage of a processing unit of the video gaming system during gameplay; and varying a precision of the weights and/or activations used in the performing MFSR, based on the monitored usage of the processing unit. A corresponding video gaming system and computer program product is also provided.