Shared Last-Level Cache Power States for Lower Energy and Fast Resume

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing computing systems waste energy by keeping a shared last level cache powered-on and active even when not in use, leading to increased power consumption and latency penalties during reinitialization.

Innovation Solution

Implement a power management policy using a data fabric and power management firmware to control the power supply to the shared last level cache, transitioning it to a retention mode during idle periods, maintaining cache content and initialization parameters, and resuming operation quickly when needed.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Speed

If the shared last level cache is kept powered-on and active, then latency is reduced and performance is maintained, but power consumption increases

Engineering Contradiction:
ImprovelatencyVSAvoidpower consumption
Core Design Contradiction:
SpeedVSUse of energy by moving object

Solution Approach 1:

The patent implements dynamic power management by transitioning the shared last level cache between different power states (fully powered-on, retention mode, and powered-off) based on actual usage conditions. This allows the system to adapt the cache's operational state to match workload demands, reducing power consumption during idle periods while maintaining quick resume capability through retention mode that preserves cache content and initialization parameters.

Inventive Principle:
Principle #15Dynamics

2Use of energy by moving object

If the shared last level cache is transitioned to retention mode during idle periods, then power consumption is reduced, but reinitialization time may increase when resuming operation

Engineering Contradiction:
Improvepower consumptionVSAvoidreinitialization time
Core Design Contradiction:
Use of energy by moving objectVSLoss of time

Solution Approach 1:

The patent applies preliminary action by maintaining cache content and initialization parameters in retention mode before complete power-down occurs. This preliminary preservation of state information allows the cache to resume operations quickly when needed, avoiding full reinitialization delays while still achieving significant power savings during extended idle periods.

Inventive Principle:
Principle #10Preliminary action

3Use of energy by moving object

If the shared last level cache is completely powered-off, then power consumption is minimized, but latency increases due to reinitialization penalties

Engineering Contradiction:
Improvepower consumptionVSAvoidlatency
Core Design Contradiction:
Use of energy by moving objectVSSpeed

Solution Approach 1:

The patent implements periodic action through a power management policy that monitors cache usage and transitions the shared last level cache between different power states based on activity detection. During idle periods, the cache enters retention mode or powered-off state to save power, while during active usage, it transitions to fully powered-on state to maintain performance, creating a periodic cycle of power state changes that balances energy efficiency with operational readiness.

Inventive Principle:
Principle #19Periodic action

Data Source

PatentUS20250217293A1Shared Last Level Cache Usage Management for Multiple Clients
Publication Date: 2025.07.03 ADVANCED MICRO DEVICES INC
  • US20250217293A1 patent drawing
  • US20250217293A1 patent drawing
  • US20250217293A1 patent drawing

AI summary

Shared last level cache usage management for multiple clients is described. In one or more implementations, a system includes a shared last level cache coupled to multiple clients and a dynamic random access memory. The system further includes a linear dropout regulator that supplies power to the shared last level cache. A data fabric included in the system is configured to control a level of the power supplied from the linear dropout regulator to be either a first level or a second level based on usage of the shared last level cache.