Multi-GPU Load Balancing via Asymmetric Frame Segmentation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing multi-processor graphics systems face challenges in efficiently distributing rendering tasks among multiple graphics processing units, leading to redundant work and suboptimal performance, making it difficult to visualize and understand the system's performance metrics such as load balancing and processor utilization.

Innovation Solution

The system employs split-frame and alternate-frame rendering modes to dynamically balance the load among multiple GPUs by dividing the display area unequally and alternating frame rendering tasks, providing real-time graphical feedback on processor efficiency and performance metrics.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If multiple processors operate in parallel, then the number of rendering operations increases, but redundant tasks are performed reducing efficiency

Engineering Contradiction:

Improvenumber of rendering operationsVSAvoidprocessor efficiency

Core Design Contradiction:

ProductivityVSLoss of energy

Solution Approach 1:

The display frame is segmented into different regions (first display region and second display region) that are assigned to different graphics processing units. This segmentation allows each processor to work on distinct portions of the rendering task, eliminating redundant operations while maintaining high productivity through parallel processing.

Inventive Principle:

Principle #1Segmentation

2Speed

If multiple processors operate in parallel, then rendering speed increases, but load distribution becomes unbalanced

Engineering Contradiction:

Improverendering speedVSAvoidload balancing

Core Design Contradiction:

SpeedVSEase of operation

Solution Approach 1:

The system dynamically adjusts the display regions assigned to each graphics processing unit based on their respective rendering speeds. When one processor is faster than the other, its display region is expanded while the slower processor's region is reduced, ensuring balanced workload distribution and maintaining optimal rendering speed across both processors.

Inventive Principle:

Principle #15Dynamics

3Ease of manufacture

If display area is divided equally among processors, then simplicity is maintained, but rendering efficiency decreases due to speed differences

Engineering Contradiction:

Improvesimplicity of divisionVSAvoidrendering efficiency

Core Design Contradiction:

Ease of manufactureVSProductivity

Solution Approach 1:

The system implements asymmetric division of the display area among graphics processing units based on their relative rendering speeds. Instead of equal division, each processor is assigned a display region proportional to its capability, with faster processors handling larger regions. This asymmetric approach maximizes rendering efficiency while the system automatically manages the complexity of dynamic region adjustment.

Inventive Principle:

Principle #4Asymmetry

Data Source

PatentUS7456833B1Graphical representation of load balancing and overlap

Publication Date: 2008.11.25 NVIDIA CORP

AI summary

Circuits, methods, and apparatus for graphically displaying performance metrics of processors such as graphics processing units in multiple processor systems. Embodiments of the present invention may provide metric information regarding operations in alternate-frame rendering, split-frame rendering, or other modes of operation. One embodiment of the present invention provides data in split-frame rendering mode including load balancing, graphics processing unit utilization, frame rate, and other types of system information in a graphical manner. Another exemplary embodiment of the present invention provides graphical information regarding graphics processing unit utilization, frame rate, and other system information while operating in the alternate-frame rendering mode.