Peripheral Component Multiplexing Bus Architecture
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional multi-GPU systems require external bridge integrated circuits (ICs) that are costly, occupy significant printed circuit board area, and consume additional power, necessitating a more efficient data transfer architecture.
Innovation Solution
The proposed solution involves a multi-processor architecture that multiplexes a peripheral bus, eliminating the need for external bridge ICs by using internal bridges within bus endpoints, allowing direct high-speed connections between GPUs and optimizing silicon design, while maintaining compatibility with existing bus protocols.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If external bridge ICs are used in multi-GPU systems, then data transfer between GPUs is enabled, but board area is consumed and cost increases
Solution Approach 1:
The patent integrates the bridge functionality directly into the GPU endpoints themselves, merging the bridge IC function with the endpoint devices. This eliminates the need for separate external bridge ICs and reduces board area by consolidating functions into existing components.
Solution Approach 2:
The patent extracts the bridge functionality from external bridge ICs and relocates it into the endpoint devices (GPUs). This extraction removes the external bridge component from the system, reducing board area while maintaining data transfer capability.
2Productivity
If external bridge ICs are used in multi-GPU systems, then data transfer between GPUs is enabled, but system cost increases
Solution Approach 1:
By merging the bridge function into the endpoint devices, the system eliminates the need for separate expensive bridge ICs. This integration reduces component count and system cost while maintaining the required data transfer functionality between GPUs.
Solution Approach 2:
The endpoint devices are designed to perform multiple functions: both as GPU endpoints and as bridges. This multi-functionality eliminates the need for dedicated bridge ICs, reducing system cost by using existing components for multiple purposes.
3Productivity
If external bridge ICs are used in multi-GPU systems, then data transfer between GPUs is enabled, but power consumption increases
Solution Approach 1:
The patent combines the bridge functionality with the endpoint devices, eliminating separate bridge ICs that would consume additional power. By integrating functions into existing components, the system reduces overall power consumption while maintaining data transfer capability.
4Area of stationary object
If the number of components is reduced by eliminating external bridges, then board area and cost are reduced, but data transfer capability must be maintained
Solution Approach 1:
The bridge functionality is extracted from external ICs and embedded into the endpoint devices. This extraction maintains data transfer capability while removing the external bridge component, thus reducing board area without sacrificing productivity.
Solution Approach 2:
Endpoint devices are designed with multi-functionality to serve both as GPUs and as bridges. This universality ensures that data transfer capability is maintained even though external bridge components have been eliminated, reducing board area while preserving functionality.
Data Source
AI summary
Embodiments of a peripheral component are described herein. Embodiments provide alternatives to the use of an external bridge integrated circuit (IC) architecture. For example, an embodiment multiplexes a peripheral bus such that multiple processors in one peripheral component can use one peripheral interface slot without requiring an external bridge IC. Embodiments are usable with known bus protocols.