GPU Deadline Scheduling to Reduce Context Switching

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing GPU scheduling methods lead to inefficient context switching and high power consumption due to prioritization-based algorithms that do not account for task deadlines and workload characteristics.

Innovation Solution

A method that estimates task completion times and compares them to deadlines, adjusting GPU operation (scheduling and power settings) to ensure deadlines are met while minimizing context switches and optimizing power usage.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If prioritization-based scheduling is used to ensure high priority tasks are executed first, then task importance is satisfied, but unnecessary context switches occur and power consumption increases

Engineering Contradiction:
Improvetask deadline meetingVSAvoidpower consumption
Core Design Contradiction:
ReliabilityVSUse of energy by moving object

Solution Approach 1:

The patent changes the scheduling parameter from static priority levels to dynamic deadline-based scheduling. The microcontroller calculates estimated completion times for GPU tasks and compares them against deadlines, adjusting scheduling decisions based on these time parameters rather than fixed priority values. This allows the system to meet deadlines while avoiding unnecessary context switches that would increase power consumption.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent implements a feedback mechanism where the microcontroller continuously monitors GPU task execution status, calculates estimated completion times, and compares them with deadlines. Based on this feedback loop, the system dynamically adjusts scheduling decisions to ensure deadline compliance while minimizing context switches and power consumption.

Inventive Principle:
Principle #23Feedback

2Reliability

If frequent context switching is performed to meet task deadlines, then task deadline compliance improves, but GPU efficiency decreases due to overhead

Engineering Contradiction:
Improvedeadline complianceVSAvoidGPU efficiency
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The patent applies preliminary action by calculating estimated completion times for GPU tasks before execution begins. The microcontroller predicts whether a task will meet its deadline and pre-determines the optimal scheduling decision, avoiding the need for frequent context switches during task execution. This preliminary assessment ensures deadline compliance while maintaining GPU efficiency.

Inventive Principle:
Principle #10Preliminary action

3Speed

If GPU operating frequency is increased to meet task deadlines, then task completion speed improves, but power consumption increases

Engineering Contradiction:
Improvetask completion speedVSAvoidpower consumption
Core Design Contradiction:
SpeedVSUse of energy by moving object

Solution Approach 1:

The patent implements dynamic frequency adjustment based on actual task execution needs. Rather than maintaining a high fixed frequency, the microcontroller monitors task progress and adjusts the GPU operating frequency dynamically - increasing it only when necessary to meet deadlines and reducing it when tasks are on track. This dynamic approach maintains task completion speed while minimizing power consumption.

Inventive Principle:
Principle #15Dynamics

Data Source

PatentEP4235421B1Controlling operation of a GPU
Publication Date: 2025.07.30 IMAGINATION TECH LTD
  • EP4235421B1 patent drawingFigure 1
  • EP4235421B1 patent drawingFigure 2
  • EP4235421B1 patent drawingFigure 3

AI summary

The operation of a GPU is controlled based on one or more deadlines by which one or more GPU tasks must be completed and estimates of the time required to complete the execution of a first GPU task (which is currently being executed) and the time required to execute one or more other GPU tasks (which are not currently being executed). Based on a comparison between the deadline(s) and the estimates, the operating parameters of the GPU may be changed.