Variable Precision Computing System for GPU Power Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Modern processors consume significant power due to high precision calculations, which is not always necessary for all functional units, especially in graphics processing units (GPUs), leading to inefficient resource usage without a substantial impact on user experience.
Innovation Solution
Implementing precision flags to selectively adjust the precision of data for different shader processing units within a GPU, allowing certain units to process data at reduced precision by zeroing specific portions of data operands, thereby reducing power consumption without compromising visual quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high precision calculations are used in functional units, then calculation accuracy is improved, but power consumption increases
Solution Approach 1:
The patent applies local quality by allowing different functional units within the GPU to operate at different precision levels. Each functional unit can be independently configured to use high precision when needed and low precision when not needed, optimizing the balance between calculation accuracy and power consumption on a local basis rather than forcing uniform precision across the entire system.
Solution Approach 2:
The patent implements dynamics by making the precision level adjustable and changeable at runtime. The system can dynamically switch between high precision and low precision modes based on the specific computational task being performed, allowing the GPU to adapt its power consumption and accuracy characteristics in real-time rather than being fixed at one precision level.
2Measurement precision
If high precision data processing is applied to all shader processing units, then processing accuracy is improved, but resource efficiency deteriorates
Solution Approach 1:
The patent enables different shader processing units to operate at different precision levels simultaneously. Some units can process data at high precision while others operate at low precision, allowing the system to optimize resource efficiency for tasks where high precision is not required while maintaining accuracy where it is needed.
Solution Approach 2:
The patent changes the precision parameter of data operands dynamically. By adjusting the precision parameter based on the specific shader program and computational requirements, the system improves resource efficiency without significantly impacting visual quality, as the precision level is tailored to match the actual needs of each processing task.
3Use of energy by moving object
If reduced precision is used to lower power consumption, then energy efficiency is improved, but visual quality may deteriorate
Solution Approach 1:
The patent changes the precision parameter of data operands based on the specific shader program being executed. By carefully selecting which functional units operate at reduced precision and which maintain high precision, the system achieves better energy efficiency while keeping the impact on visual quality minimal or imperceptible.
Solution Approach 2:
The patent applies reduced precision selectively to specific functional units and data paths where it has minimal impact on visual quality, while maintaining high precision in other areas. This localized application of precision reduction allows the system to improve energy efficiency without significantly degrading the overall visual output quality.
Data Source
AI summary
A processor selectively adjusts the precision of data for different functional units. Specified functional units of the processor, such as shader processing unit of a graphics processing unit (GPU) include a zeroing module to store, based on the states of corresponding precision flags, a data value of zero at specified portion of an input and/or output data operand. The functional unit then processes the data including the zeroed portion. Because a portion of the data has been zeroed, the functional unit consumes less power during data processing. Furthermore, the precision flags are set such that the reduced precision of the data does not significantly impact a user experience.