Hardware-Software Power Budget Interface for Multi-Thread Performance

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing power management solutions for computing systems, particularly in high-performance computing environments, are sub-optimal in handling performance variations across collaborating threads, leading to inefficiencies in power and thermal management, and lack a unified and scalable approach to express dynamic power and performance tradeoffs.

Innovation Solution

A hardware-software interface is introduced that enables dynamic power allocation across processing nodes, allowing for flexible and efficient control of power and performance across arbitrary groups of computing threads and resources, using a 'universal coin' for deterministic behavior and a centralized control interface with scalable granularity, managed through power control and measurement interfaces.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Use of energy by moving object

If OS-based mechanisms (ACPI) are used to control processor performance states, then power consumption is reduced, but performance optimization for multiple collaborating threads is sub-optimal

Engineering Contradiction:

Improvepower consumptionVSAvoidperformance optimization for multiple threads

Core Design Contradiction:

Use of energy by moving objectVSProductivity

Solution Approach 1:

A hardware performance state controller is introduced as an intermediary between the operating system and processor cores. This controller receives performance state requests from the OS and translates them into appropriate processor states, enabling coordinated control of multiple collaborating threads while maintaining power efficiency. The hardware mediator resolves the conflict between power savings and multi-thread performance optimization.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Speed

If processor operates at higher performance states (P0) to maximize speed, then execution time is reduced, but power consumption and thermal output increase

Engineering Contradiction:

Improveprocessor execution speedVSAvoidpower consumption

Core Design Contradiction:

SpeedVSUse of energy by moving object

Solution Approach 1:

The system dynamically adjusts processor performance states based on real-time workload requirements and power availability. The hardware controller continuously monitors collaboration status of threads and transitions processors between performance states (P0-PN) accordingly, optimizing the balance between execution speed and power consumption for each specific operational context.

Inventive Principle:

Principle #15Dynamics

3Adaptability or versatility

If power management is controlled solely by software (OS), then flexibility is improved, but control precision and response time are limited

Engineering Contradiction:

Improvesoftware control flexibilityVSAvoidpower control precision

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The system merges software-based power management flexibility with hardware-based control precision. The OS maintains high-level power policy decisions while a hardware performance state controller executes precise frequency and voltage adjustments. This combination achieves both software adaptability and hardware-level control precision for power management.

Inventive Principle:

Principle #5Merging (Combining)

Data Source

PatentUS11650652B2Systems and methods for synergistic software-hardware power budget management

Publication Date: 2023.05.16 INTEL CORP

AI summary

Apparatus, systems, and methods provide an interface between a plurality of hardware resources of a node and a power manager. The interface is configured to define one or more resource groups to expose to the power manager for power measurement and control, assign the plurality of hardware resources to the one or more resource groups, and provide a power allowance to each resource group.