FPGA AFU Thermal Management via BMC Dynamic Disabling

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Information handling systems with field-programmable gate arrays (FPGAs) face challenges in power management and load balancing, particularly in thermal management and optimizing AFU placement, as existing methods can disrupt processing tasks and consume excessive system resources.

Innovation Solution

A system and method utilizing a baseboard management controller (BMC) to dynamically manage and reprogram accelerated function units (AFUs) on FPGA add-in cards, including disabling AFUs to reduce power consumption and optimize placement based on thermal thresholds and system configurations, allowing for efficient processing task redistribution without interrupting hosted environments.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Use of energy by moving object

If existing power management methods are used for FPGAs, then system resources are consumed, but processing tasks are disrupted and power consumption is excessive

Engineering Contradiction:
Improvepower consumptionVSAvoidprocessing task continuity
Core Design Contradiction:
Use of energy by moving objectVSReliability

Solution Approach 1:

The patent introduces a Baseboard Management Controller (BMC) as an intermediary device to manage FPGA power consumption. The BMC monitors power usage and dynamically reconfigures the FPGA without interrupting processing tasks, thereby resolving the contradiction between reducing power consumption and maintaining task continuity. The BMC acts as a mediator between the FPGA and the system, enabling seamless power management.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Use of energy by moving object

If dynamic reconfiguration of AFUs is implemented, then power management is improved, but system complexity increases

Engineering Contradiction:
Improvepower management efficiencyVSAvoidsystem architecture complexity
Core Design Contradiction:
Use of energy by moving objectVSDevice complexity

Solution Approach 1:

The patent implements self-service mechanisms where the FPGA monitors its own power consumption and automatically reconfigures its Accelerated Function Units (AFUs) based on predefined power thresholds. This self-monitoring and self-reconfiguration capability reduces the need for complex external control systems, thereby improving power management efficiency while minimizing the increase in system complexity.

Inventive Principle:
Principle #25Self-service

3Temperature

If thermal management is implemented by disabling AFUs, then temperature is reduced, but processing productivity decreases

Engineering Contradiction:
ImproveFPGA temperatureVSAvoidprocessing throughput
Core Design Contradiction:
TemperatureVSProductivity

Solution Approach 1:

The patent employs dynamic thermal management where the FPGA continuously monitors its temperature and dynamically adjusts the configuration of AFUs in real-time. When temperature exceeds thresholds, specific AFUs are disabled or reconfigured to reduce heat generation. When temperature returns to acceptable levels, AFUs are reactivated. This dynamic approach maintains productivity by only disabling AFUs when necessary, rather than permanently reducing processing capacity.

Inventive Principle:
Principle #15Dynamics

4Productivity

If load balancing of personality bitstreams is implemented, then processing efficiency is optimized, but BMC resource consumption increases

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

Solution Approach 1:

The patent implements preliminary action by pre-configuring multiple personality bitstreams in the BMC's memory before they are needed. The BMC loads and stores these bitstream configurations in advance, allowing for rapid switching between different FPGA configurations without requiring extensive real-time processing. This preliminary preparation optimizes processing efficiency while minimizing the BMC's real-time energy consumption during actual load balancing operations.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentUS11256314B2System and method for power management of field-programmable gate arrays and load balancing of personality bitstreams from a baseboard management controller
Publication Date: 2022.02.22 DELL PROD LP
  • US11256314B2 patent drawing
  • US11256314B2 patent drawing
  • US11256314B2 patent drawing

AI summary

An information handling system includes a processor, a system baseboard management controller (BMC), and a field-programmable gate array (FPGA) add-in card. The FPGA add-in card includes an FPGA programmed with accelerated function units (AFUs) to perform processing tasks for the processor. The AFUs include AFUs of a common type. A card BMC provides a temperature indication to the system BMC. The system BMC determines that a temperature of the FPGA add-in card exceeds a temperature threshold based upon the temperature indication, selects one of the common AFUs to be disabled, and directs the card BMC to disable the selected AFU. The card BMC disables the first AFU and not the second AFU in response to the direction to disable the first AFU.