Multi-cluster processor dynamic core activation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Multi-cluster processing systems face high power consumption due to the continuous supply of power to all cores during operation, which is inefficient and excessive.

Innovation Solution

A multi-cluster processing system with a control unit that dynamically enables or disables clusters and cores based on the number of threads to be processed and core utilization, allowing only the required number of cores to operate, thereby reducing power consumption.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If power is supplied to all multi-cores included in the cluster during operation, then processing capacity is maximized, but power consumption becomes excessive

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

Solution Approach 1:

The multi-core processor is divided into multiple clusters, where each cluster contains a subset of cores. This segmentation allows the system to activate only specific clusters containing the required number of cores for the current task, rather than powering all cores simultaneously. The control unit manages cluster activation based on thread requirements, enabling selective power supply to reduce overall power consumption while maintaining necessary processing capacity.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system dynamically adjusts the number of active cores and clusters based on real-time processing requirements. The control unit monitors the number of threads to be processed and core utilization, then dynamically enables or disables clusters and cores accordingly. This dynamic adaptation allows the system to optimize the balance between processing capacity and power consumption for varying workloads.

Inventive Principle:
Principle #15Dynamics

2Reliability

If all multi-cores are continuously powered, then system performance is maintained, but power waste increases

Engineering Contradiction:
Improvesystem performanceVSAvoidpower waste
Core Design Contradiction:
ReliabilityVSLoss of energy

Solution Approach 1:

The system changes operational parameters by adjusting the number of active cores and clusters based on processing needs. Instead of maintaining all cores in an active state, the control unit modifies the activation state of clusters and cores according to the number of threads requiring processing. This parameter adjustment ensures system performance is maintained at the necessary level while eliminating power waste from idle or underutilized cores.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The control unit autonomously manages cluster and core activation without requiring external intervention. It self-monitors processing requirements and core utilization, then automatically enables or disables appropriate clusters and cores. This self-service mechanism ensures system performance requirements are met while minimizing power consumption through intelligent, autonomous resource management.

Inventive Principle:
Principle #25Self-service

Data Source

PatentEP2674863B1Multi-cluster processing system and method of operating the same
Publication Date: 2020.05.27 SAMSUNG ELECTRONICS CO LTD
  • EP2674863B1 patent drawingFigure 1
  • EP2674863B1 patent drawingFigure 2
  • EP2674863B1 patent drawingFigure 3

AI summary

A multi-cluster processing system and a method of operating a multi-cluster processing system are provided. The multi-cluster processing system includes: a first cluster including a plurality of first-type cores; a second cluster including a plurality of second-type cores; and a control unit configured to monitor loads of the first-type cores and the second-type cores, wherein when utilization of at least one of enabled first-type cores exceeds a predetermined threshold utilization of each of the first-type cores, the control unit enables at least one of disabled first-type cores in a first mode, and the control unit enables at least one of the disabled second-type cores and disables the first cluster in a second mode, wherein an amount of computation per unit of time of each of the second-type cores is greater than an amount of computation per unit of time of each of the first-type cores.