Dynamic SIMD Deactivation for GPU Power Reduction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional graphics processors waste power by keeping all SIMDs active, even when underutilized, due to the inability to dynamically control and disable individual SIMDs, leading to inefficient power consumption and performance per watt.

Innovation Solution

Implementing dynamic activation and deactivation of individual SIMDs in a shader complex, using load balancing per watt techniques and dynamic grain clock gating, to only activate SIMDs needed for the workload, reducing power consumption and improving performance per watt without compromising system performance.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Power

If all SIMDs are kept enabled to maintain system performance, then processing capability is preserved, but power consumption increases

Engineering Contradiction:
Improvepower consumptionVSAvoidprocessing capability
Core Design Contradiction:
PowerVSProductivity

Solution Approach 1:

The patent implements dynamic enabling and disabling of SIMD units based on real-time workload assessment. The shader engine controller monitors workload conditions and dynamically activates or deactivates individual SIMDs, transitioning from static configuration to adaptive dynamic control, thereby optimizing the balance between power consumption and processing capability

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The system changes the operational state parameter of SIMD units from a fixed enabled/disabled state to a dynamically adjustable state based on workload intensity. By monitoring workload conditions and adjusting SIMD activation status in real-time, the system adapts power consumption and processing capability parameters to match actual computational demands

Inventive Principle:
Principle #35Parameter changes

2Loss of energy

If conventional static enabling methods are used, then system stability is maintained, but power efficiency deteriorates

Engineering Contradiction:
Improvepower efficiencyVSAvoidsystem stability
Core Design Contradiction:
Loss of energyVSReliability

Solution Approach 1:

The patent implements a feedback mechanism where the shader engine controller continuously monitors workload conditions and uses this information to dynamically adjust SIMD activation status. This closed-loop control system provides feedback from actual computational demands to the configuration controller, enabling adaptive optimization of power efficiency while maintaining system stability through controlled transitions

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The shader engine controller autonomously monitors workload conditions and self-adjusts SIMD configuration without external intervention. The system services its own configuration needs by automatically assessing computational demands and enabling or disabling appropriate SIMD units, thereby improving power efficiency while maintaining operational stability

Inventive Principle:
Principle #25Self-service

3Productivity

If workload is distributed across all SIMDs, then processing throughput is maintained, but power consumption increases

Engineering Contradiction:
Improveprocessing throughputVSAvoidpower consumption
Core Design Contradiction:
ProductivityVSUse of energy by moving object

Solution Approach 1:

The patent applies local quality by enabling or disabling specific SIMD units based on their individual workload requirements rather than uniformly managing all SIMDs. The controller assesses local workload conditions for different SIMD units and configures each accordingly, allowing some SIMDs to remain active while others are deactivated, thereby reducing overall power consumption while maintaining necessary processing throughput

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The system segments the SIMD array into independently controllable units that can be individually enabled or disabled based on workload assessment. By dividing the processing resources into separable, independently manageable segments, the system can activate only the necessary portion of SIMDs for current computational tasks, reducing power consumption while preserving required processing throughput

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP2593860B1Dynamic enabling and disabling of SIMD units in a graphics processor
Publication Date: 2020.08.19 ADVANCED MICRO DEVICES INC
  • EP2593860B1 patent drawingFigure 1A
  • EP2593860B1 patent drawingFigure 1B
  • EP2593860B1 patent drawingFigure 1C

AI summary

Systems and methods to improve performance in a graphics processing unit are described herein. Embodiments achieve power saving in a graphics processing unit by dynamically activating/deactivating individual SIMDs in a shader complex that comprises multiple SIMD units. On-the-fly dynamic disabling and enabling of individual SIMDs provides flexibility in achieving a required performance and power level for a given processing application. In this way, embodiments achieve optimal usage of a graphics processing unit. Embodiments of the invention also achieve dynamic grain (e.g., medium grain) clock gating of SIMDs in a shader complex. Embodiments reduce switching power by shutting down clock trees to unused logic by providing a clock on demand mechanism. In this way, embodiments enhance clock gating to save more switching power for the duration of time when SIMDs are idle (or assigned no work).