Workload-Aware GPU Execution Unit Powerdown via Granular DVFS

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

The increasing power requirements of graphics processors strain the power delivery capability of consumer computing systems, necessitating efficient power management strategies to balance performance and energy consumption.

Innovation Solution

A chiplet architecture with granular dynamic voltage and frequency scaling (DVFS) is implemented, allowing for power delivery configuration on a per-chunk basis, grouping homogenous or heterogeneous chiplets into chunks with uniform power requirements, and employing a chunking architecture to optimize power distribution.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Power

If the amount of logic on integrated circuit devices is increased to improve processing power, then computing performance is improved, but power requirements escalate and stress power delivery capability

Engineering Contradiction:
Improveprocessing powerVSAvoidpower requirements
Core Design Contradiction:
PowerVSUse of energy by moving object

Solution Approach 1:

The integrated circuit is divided into multiple voltage domains, each independently regulatable. This segmentation allows different regions to receive appropriate power levels based on their specific needs, improving overall power efficiency while maintaining high processing power where required.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

Different voltage domains are assigned different power delivery capabilities based on local requirements. High-performance regions receive higher power, while less demanding regions receive lower power, optimizing the balance between processing power and energy consumption across the chip.

Inventive Principle:
Principle #3Local quality

2Adaptability or versatility

If power delivery is configured on a per-chunk basis with granular DVFS, then power management efficiency is improved, but device complexity increases

Engineering Contradiction:
Improvepower management efficiencyVSAvoiddevice complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The device is divided into chunks of voltage domains that can be independently controlled. This segmentation enables granular power management where each chunk can be scaled independently, improving adaptability while the modular structure helps manage complexity through systematic organization.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

Dynamic voltage and frequency scaling is implemented at the chunk level, allowing the system to adapt power delivery in real-time based on workload requirements. This dynamic control improves power management efficiency while the automated nature of DVFS helps mitigate the complexity burden.

Inventive Principle:
Principle #15Dynamics

Data Source

PatentUS12461585B2Granular GPU DVFS with execution unit partial powerdown
Publication Date: 2025.11.04 INTEL CORP
  • US12461585B2 patent drawing
  • US12461585B2 patent drawing
  • US12461585B2 patent drawing

AI summary

Described herein, in one embodiment, are techniques to facilitate the partial powerdown of sub-components of an execution unit or other graphics processing resource based on the workload to be executed. In another embodiment, granular dynamic voltage and frequency scaling is enabled in which the voltage and frequency of groups of processing resources within a graphics processor can be separately scaled.