Fixed-Point Alpha Blending Architecture for GPU Power and Die Area

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current graphics processing units (GPUs) face challenges in optimizing alpha blending operations due to the need for dedicated floating-point hardware, which consumes excessive power and die area, and often provides higher precision than required for lower precision color buffer formats, leading to inefficiencies in power management and heat generation.

Innovation Solution

A system that utilizes both fixed-point and floating-point blending units, with a multiplexer to combine their outputs, allowing for efficient processing of alpha channel data using a processor and API to route data streams through either representation as needed, thereby optimizing power consumption and reducing hardware complexity.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If dedicated floating-point hardware is used for blending operations, then precision is improved, but power consumption and die area increase

Engineering Contradiction:
Improveblending precisionVSAvoidpower consumption
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The system dynamically selects between fixed-point and floating-point blending units based on the precision requirements of the current blending operation. The API determines whether to route alpha channel data to the fixed-point blending unit or the floating-point blending unit, allowing the system to adapt its precision level to match the actual needs of each operation, thereby avoiding unnecessary power consumption from using floating-point hardware when fixed-point suffices.

Inventive Principle:
Principle #15Dynamics

2Measurement precision

If dedicated floating-point hardware is used for blending operations, then precision is improved, but die area increases

Engineering Contradiction:
Improveblending precisionVSAvoiddie area
Core Design Contradiction:
Measurement precisionVSArea of stationary object

Solution Approach 1:

The system implements a universal blending architecture that includes both fixed-point and floating-point blending units, allowing the same hardware infrastructure to serve multiple precision requirements. The API routes blending operations to the appropriate unit based on needs, making the floating-point unit optional and shareable across multiple functions, thereby reducing the overall die area compared to having dedicated floating-point hardware for all blending operations.

Inventive Principle:
Principle #6Universality (Multi-functionality)

3Use of energy by moving object

If fixed-point blending unit is used, then power consumption is reduced, but precision may be insufficient for high-precision operations

Engineering Contradiction:
Improvepower consumptionVSAvoidblending precision
Core Design Contradiction:
Use of energy by moving objectVSMeasurement precision

Solution Approach 1:

The API acts as an intermediary that intelligently routes alpha channel data to either the fixed-point or floating-point blending unit based on the specific operation requirements. For operations where high precision is needed, the API directs data to the floating-point unit; for operations where lower precision suffices, it uses the fixed-point unit, thus optimizing the balance between power consumption and precision.

Inventive Principle:
Principle #24Intermediary (Mediator)

4Measurement precision

If floating-point hardware is used, then precision is improved, but heat generation increases

Engineering Contradiction:
Improveblending precisionVSAvoidheat generation
Core Design Contradiction:
Measurement precisionVSTemperature

Solution Approach 1:

The system changes the precision parameter dynamically by selecting between fixed-point and floating-point representations based on the operational requirements. This parameter change allows the system to use the lower-power fixed-point mode for routine operations, thereby reducing heat generation, while reserving floating-point mode for specific high-precision tasks that truly require it.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP3392838B1Dedicated fixed point blending for energy consumption efficiency
Publication Date: 2021.12.15 INTEL CORP
  • EP3392838B1 patent drawingFigure 1
  • EP3392838B1 patent drawingFigure 2A
  • EP3392838B1 patent drawingFigure 2B

AI summary

Briefly, in accordance with one or more embodiments, a processor receives an incoming data stream that includes alpha channel data, and a memory stores an application programming interface (API). The API is to route the alpha channel data to a fixed point blending unit to perform one or more blending operations using fixed point representation of the alpha channel data. The API is further to route the incoming data stream to a floating point blending unit to perform operations involving floating point representation of the incoming data.