Adaptive Power Management in Accelerator Sleds

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Data centers face challenges in efficiently managing power utilization and thermal limitations when using multiple accelerator devices, leading to increased costs due to strain on circuit boards and potential over-performance that exceeds service level agreement requirements.

Innovation Solution

The implementation of a disaggregated resource architecture with chassis-less sleds that allow for independent upgrading and management of resources, including adaptive power management logic to optimize power usage based on predicted patterns and service level agreements, ensuring efficient operation within defined power budgets.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If full power is provided to all accelerator devices on the circuit board, then the processing speed and performance of accelerator devices are improved, but the thermal and electrical limitations of the circuit board are strained and cooling expenses increase

Engineering Contradiction:

Improveprocessing speedVSAvoidthermal limitations

Core Design Contradiction:

ProductivityVSTemperature

Solution Approach 1:

The patent implements dynamic power management by continuously monitoring performance metrics and adjusting power allocation to accelerator devices in real-time. The system transitions from static full-power operation to dynamic adaptive power levels, modifying power delivery based on actual workload demands and thermal conditions to resolve the contradiction between maintaining high processing speed and managing thermal limitations.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The system changes operational parameters by adjusting power allocation levels to accelerator devices based on monitored performance data. By varying power parameters dynamically rather than maintaining constant full-power operation, the system achieves optimal processing speed while staying within thermal and electrical limitations of the circuit board.

Inventive Principle:

Principle #35Parameter changes

2Productivity

If full power is provided to all accelerator devices on the circuit board, then the processing speed and performance of accelerator devices are improved, but cooling expenses and equipment replacement costs increase

Engineering Contradiction:

Improveprocessing speedVSAvoidcooling expenses

Core Design Contradiction:

ProductivityVSLoss of energy

Solution Approach 1:

The system implements dynamic power management by continuously monitoring performance metrics and adjusting power allocation to accelerator devices in real-time. The system transitions from static full-power operation to dynamic adaptive power levels, modifying power delivery based on actual workload demands and thermal conditions to resolve the contradiction between maintaining high processing speed and managing thermal limitations.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

Inventive Principle:

Principle #35Parameter changes

3Loss of energy

If adaptive power management is implemented, then power utilization efficiency is improved and costs are reduced, but system complexity increases

Engineering Contradiction:

Improvepower utilization efficiencyVSAvoidsystem complexity

Core Design Contradiction:

Loss of energyVSDevice complexity

Solution Approach 1:

The patent implements self-service power management where the system automatically monitors its own performance metrics, predicts future power needs using machine learning algorithms, and adjusts power allocation without external intervention. This autonomous operation reduces the need for complex external control systems while improving power utilization efficiency through continuous adaptive optimization.

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The system incorporates feedback mechanisms by continuously monitoring performance metrics from accelerator devices and using this data to adjust power allocation. The feedback loop includes performance monitoring, prediction using machine learning, and power adjustment, creating a closed-loop control system that optimizes power efficiency while managing complexity through automated decision-making.

Inventive Principle:

Principle #23Feedback

Data Source

PatentEP3731063B1Technologies for providing adaptive power management in an accelerator sled

Publication Date: 2022.10.05 INTEL CORP

Figure 1
Figure 2
Figure 3

AI summary

Technologies for providing adaptive power management in an accelerator sled include an accelerator sled having circuitry to determine, based on (i) a total power budget for the accelerator sled, (ii) service level agreement (SLA) data indicative of a target performance of a kernel, and (iii) profile data indicative of a performance of the kernel as a function of a power utilization of the kernel, a power utilization limit for the kernel to be executed by an accelerator device on the accelerator sled. Additionally, the circuitry is to allocate the determined power utilization limit to the kernel and execute the kernel under the allocated power utilization limit.