Distributed GPU Architecture for Near-Storage Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current GPU architectures are limited in their memory storage capabilities and processing efficiency, particularly in handling large datasets and performing near-storage processing.
Innovation Solution
A distributed graphics processor unit (GPU) architecture that consists of an array of processing nodes, each equipped with a GPU node, fast memory, and storage unit, allowing for near-storage processing and improved memory storage capabilities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If GPUs use bandwidth processors with small capacity to interface with host memory, then processing speed is improved, but memory storage capability deteriorates
Solution Approach 1:
The system is divided into multiple processing nodes, each with its own GPU, fast memory, and storage unit. This segmentation allows each node to independently process data locally while collectively providing large storage capacity and high processing throughput across the distributed system.
Solution Approach 2:
The patent transitions from a single-GPU architecture to a distributed array of processing nodes, adding a dimensional aspect to the system. This distributed architecture enables the system to achieve both high processing speed at the node level and large storage capacity at the system level by operating across multiple spatial dimensions.
2Device complexity
If data is processed remotely from storage location, then system simplicity is maintained, but data traffic and processing efficiency deteriorate
Solution Approach 1:
Each processing node is equipped with local fast memory and storage units that are directly coupled to its GPU. This local quality enables data to be processed close to where it is stored, reducing the need for data movement across the system and improving processing efficiency without significantly increasing overall system complexity.
3Device complexity
If single chip GPU architecture is used, then device simplicity is maintained, but aggregate bandwidth and throughput deteriorate
Solution Approach 1:
The patent merges multiple processing nodes into a unified distributed system where each node contains a GPU, fast memory, and storage unit. By combining the resources of multiple nodes, the system achieves high aggregate bandwidth and multi-teraflop throughput while maintaining relative simplicity at the individual node level.
Data Source
AI summary
The present disclosure is directed to a distributed graphics processor unit (GPU) architecture that includes an array of processing nodes. Each processing node may include a GPU node that is coupled to its own fast memory unit and its own storage unit. The fast memory unit and storage unit may be integrated into a single unit or may be separately coupled to the GPU node. The processing node may have its fast memory unit coupled to both the GPU node and the storage node. The various architectures provide a GPU-based system that may be treated as a storage unit, such as solid state drive (SSD) that performs onboard processing to perform memory-oriented operations. In this respect, the system may be viewed as a “smart drive” for big-data near-storage processing.