Hybrid GPU Data Fabric for Scalable Multi-Core Communication

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Multi-core graphics processing units face scalability issues due to communication solutions that are not suitable for both low-power and high-performance segments, leading to inefficiencies in power consumption and performance.

Innovation Solution

A data distribution fabric is introduced for scalable graphics processors, enabling efficient communication between graphics cores by using a data distribution fabric that supports multiple power and performance segments, optimizing power consumption and performance.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Speed

If unique data distribution channels are used to couple graphics processors, then communication performance is improved, but device complexity and cost increase

Engineering Contradiction:
Improvecommunication performanceVSAvoidcommunication solution complexity
Core Design Contradiction:
SpeedVSDevice complexity

Solution Approach 1:

The patent implements a universal communication fabric that serves multiple functions: it provides high-performance communication for power-intensive graphics processors while also supporting low-power graphics processors with reduced functionality. The same fabric structure accommodates different performance segments through configurable routing and resource allocation, eliminating the need for separate communication channels for different power segments.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The communication fabric employs dynamic resource allocation and configurable routing that adapts to different power and performance segments. The system can dynamically adjust communication pathways, bandwidth allocation, and routing tables based on the specific needs of each graphics processor segment, allowing optimal performance for high-power units while maintaining efficiency for low-power units without requiring dedicated fixed channels.

Inventive Principle:
Principle #15Dynamics

2Productivity

If high-end multi-core communication solutions are used, then graphics processing performance is improved, but power consumption increases

Engineering Contradiction:
Improvegraphics processing performanceVSAvoidpower consumption
Core Design Contradiction:
ProductivityVSUse of energy by moving object

Solution Approach 1:

The patent applies local quality by providing different communication capabilities and power management features to different segments of the graphics processor. High-power segments receive enhanced communication resources and performance optimizations, while low-power segments use simplified communication pathways and power management modes. This localized differentiation allows each segment to operate with appropriate resources, avoiding unnecessary power consumption in low-power segments while maintaining high performance where needed.

Inventive Principle:
Principle #3Local quality

3Use of energy by moving object

If low-power graphics communication solutions are used, then power consumption is reduced, but graphics processing performance deteriorates

Engineering Contradiction:
Improvepower consumptionVSAvoidgraphics processing performance
Core Design Contradiction:
Use of energy by moving objectVSProductivity

Solution Approach 1:

The patent segments the graphics processor into different power and performance segments, each with tailored communication and processing capabilities. This segmentation allows low-power segments to operate efficiently with reduced power consumption while high-power segments maintain high performance. The segmented architecture prevents the performance bottleneck that would occur if the entire system was limited to low-power operation, as each segment can operate at its optimal performance level.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP4283950B1Data distribution fabric in scalable gpus
Publication Date: 2026.04.15 INTEL CORP
  • EP4283950B1 patent drawingFigure 1
  • EP4283950B1 patent drawingFigure 2
  • EP4283950B1 patent drawingFigure 3

AI summary

In on embodiment, a hybrid fabric interconnects multiple graphics processor cores within a processor. The hybrid fabric interconnect includes multiple data channels, including programmable virtual data channels. The virtual data channels carry multiple traffic classes of packet-based messages. The virtual data channels and multiple traffic classes may be assigned one of multiple priorities. The virtual data channels may be arbitrated independently. The hybrid fabric is scalable and can support multiple topologies, including multiple stacked integrated circuit topologies.