Reconfigurable CPU-GPU Interconnect for Thermal Throttling

Solution Overview

Problem

High-speed workload processing architectures face challenges with power/thermal throttling in rack-mount server systems, leading to thermal hot spots and excessive power conduction, which impact performance and reliability.

Innovation Solution

A re-configurable system architecture that dynamically adjusts the network topology by reconfiguring cable connections between CPUs and GPUs, using heat sensors and a chassis management module to monitor and mitigate thermal hot spots and power issues, allowing for optimal workload distribution and cooling.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If high-speed workload processing is implemented using multiple CPUs and GPUs in rack-mount systems, then processing throughput is improved, but thermal hot spots and power conduction issues occur

Engineering Contradiction:

Improveworkload processing throughputVSAvoidthermal hot spots

Core Design Contradiction:

ProductivityVSTemperature

Solution Approach 1:

The system dynamically reconfigures the network topology by changing cable connections between CPUs and GPUs based on real-time thermal conditions. The chassis management module monitors temperature and power consumption, then adjusts the interconnect topology to redistribute workload and heat generation across different processing units, transforming a static architecture into a dynamic adaptive system that responds to thermal conditions.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

Different regions of the rack-mount system are treated with different quality characteristics. The system identifies specific thermal hot spots and applies localized cooling strategies while redistributing workload away from overheated regions. Each computing node can have different workload assignments and connectivity configurations based on its thermal state, creating non-uniform local conditions to manage overall system temperature.

Inventive Principle:

Principle #3Local quality

2Productivity

If high-speed workload processing is implemented using multiple CPUs and GPUs in rack-mount systems, then processing throughput is improved, but excessive power conduction occurs

Engineering Contradiction:

Improveworkload processing throughputVSAvoidpower conduction

Core Design Contradiction:

ProductivityVSUse of energy by stationary object

Solution Approach 1:

The system dynamically reconfigures the network topology by changing cable connections between CPUs and GPUs based on real-time thermal conditions. The chassis management module monitors temperature and power consumption, then adjusts the interconnect topology to redistribute workload and heat generation across different processing units, transforming a static architecture into a dynamic adaptive system that responds to thermal conditions.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The chassis management module implements a feedback loop by continuously monitoring power consumption and thermal conditions, then using this information to make intelligent decisions about workload distribution and topology reconfiguration. The system measures power conduction levels, compares them against thresholds, and adjusts resource allocation accordingly, creating a closed-loop control system that optimizes energy efficiency.

Inventive Principle:

Principle #23Feedback

3Device complexity

If fixed network topology is used for GPU-CPU interconnection, then system complexity is reduced, but ability to mitigate thermal hot spots is limited

Engineering Contradiction:

Improvenetwork topology configurationVSAvoidthermal management effectiveness

Core Design Contradiction:

Device complexityVSReliability

Solution Approach 1:

The system dynamically reconfigures the network topology by changing cable connections between CPUs and GPUs based on real-time thermal conditions. The chassis management module monitors temperature and power consumption, then adjusts the interconnect topology to redistribute workload and heat generation across different processing units, transforming a static architecture into a dynamic adaptive system that responds to thermal conditions.

Inventive Principle:

Principle #15Dynamics

Data Source

PatentUS10747280B2Reconfigurble CPU/GPU interconnect to mitigate power/thermal throttling

Publication Date: 2020.08.18 INTERNATIONAL BUSINESS MACHINE CORPORATION

AI summary

A method, a system and a computer program product for reconfiguring hardware network topology including graphics processor units (GPU) and central processing unit (CPU) interconnectivity on or across compute nodes of a rack-mount server. The re-configurability is based on detected thermal throttling or thermal hot spots when running workloads. For network re-configurability, a user can directly connect high-speed cable links between CPU/GPU connectors and between GPU/GPU connectors on a same PCB compute node, or across two PCB compute nodes as suggested by a control processor to avoid thermal and power hotspots when running the workload. The method recommends and generates a system map of the hardware network topology known to avoid/mitigate thermal throttling, and instructs a configuration of CPUs and GPUs such that GPUs are assigned to workloads at locations for mitigating thermal throttling based on detected thermal hot spots and power hot spots to optimize workload performance.