Thermal Aware Thread Distribution for GPU Compute Units

Solution Overview

Problem

Graphics Processing Units (GPUs) face thermal issues due to uneven distribution of threads across compute units, leading to performance throttling and overheating, as certain units have higher thermal costs proximity to high temperature regions.

Innovation Solution

Implementing thermal aware optimization logic to distribute threads based on thermal feedback, location of heat sources, and wavefront characteristics, selecting compute units that minimize temperature and maximize performance by reducing thermal throttling.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of operation

If threads are distributed to compute units on a first come, first serve basis, then thread execution simplicity is maintained, but thermal costs increase and performance throttling occurs

Engineering Contradiction:

Improvethread distribution simplicityVSAvoidGPU temperature

Core Design Contradiction:

Ease of operationVSTemperature

Solution Approach 1:

The patent applies local quality by assigning different scheduling policies to different compute units based on their thermal characteristics. Compute units are divided into groups (e.g., hot spots vs. cooler regions), and threads are selectively assigned to different groups depending on wavefront characteristics and thermal feedback, thereby optimizing thermal distribution while maintaining performance.

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The patent implements feedback mechanisms where thermal information from compute units is continuously monitored and used to adjust thread scheduling decisions. The scheduler receives thermal feedback and modifies its assignments dynamically, balancing workload distribution to prevent thermal throttling while maintaining efficient thread execution.

Inventive Principle:

Principle #23Feedback

2Productivity

If threads are assigned to any available idle compute unit, then scheduling speed is improved, but thermal throttling occurs due to uneven thermal distribution

Engineering Contradiction:

Improvethread execution throughputVSAvoidthermal stability

Core Design Contradiction:

ProductivityVSReliability

Solution Approach 1:

The patent applies local quality by assigning different scheduling policies to different compute units based on their thermal characteristics. Compute units are divided into groups (e.g., hot spots vs. cooler regions), and threads are selectively assigned to different groups depending on wavefront characteristics and thermal feedback, thereby optimizing thermal distribution while maintaining performance.

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The patent implements feedback mechanisms where thermal information from compute units is continuously monitored and used to adjust thread scheduling decisions. The scheduler receives thermal feedback and modifies its assignments dynamically, balancing workload distribution to prevent thermal throttling while maintaining efficient thread execution.

Inventive Principle:

Principle #23Feedback

3Device complexity

If compute units are located closer together to increase density, then device integration is improved, but thermal management becomes more difficult

Engineering Contradiction:

Improvecompute unit densityVSAvoidthermal distribution uniformity

Core Design Contradiction:

Device complexityVSTemperature

Solution Approach 1:

The patent applies local quality by assigning different scheduling policies to different compute units based on their thermal characteristics. Compute units are divided into groups (e.g., hot spots vs. cooler regions), and threads are selectively assigned to different groups depending on wavefront characteristics and thermal feedback, thereby optimizing thermal distribution while maintaining performance.

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The patent segments compute units into different groups or zones based on their thermal characteristics and proximity to heat sources. This segmentation allows the scheduling system to treat different regions differently, assigning threads to cooler regions when thermal feedback indicates high temperatures in other areas, thereby managing thermal distribution in high-density architectures.

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS11726837B2Apparatus and method for providing workload distribution of threads among multiple compute units

Publication Date: 2023.08.15 ADVANCED MICRO DEVICES INC

AI summary

In some examples, thermal aware optimization logic determines a characteristic (e.g., a workload or type) of a wavefront (e.g., multiple threads). For example, the characteristic indicates whether the wavefront is compute intensive, memory intensive, mixed, and/or another type of wavefront. The thermal aware optimization logic determines temperature information for one or more compute units (CUs) in one or more processing cores. The temperature information includes predictive thermal information indicating expected temperatures corresponding to the one or more CUs and historical thermal information indicating current or past thermal temperatures of at least a portion of a graphics processing unit (GPU). The logic selects the one or more compute units to process the plurality of threads based on the determined characteristic and the temperature information. The logic provides instructions to the selected subset of the plurality of CUs to execute the wavefront.