Accelerator-Based Data Computing for Lower Bus Occupation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional computing methods in AI technology result in high bus resource occupation due to frequent processor memory access, adversely affecting execution efficiency.
Innovation Solution
A data computing system with an accelerator that performs computing independently, reducing processor overhead and bus resource usage by offloading data processing tasks to the accelerator, allowing the processor to handle other events.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the processor accesses memory constantly during computing, then data processing can be performed, but bus resource occupation increases and execution efficiency deteriorates
Solution Approach 1:
The system segments computing tasks into two parts: control functions remain with the processor, while data computing is offloaded to the accelerator. This segmentation allows the processor to avoid constant memory access for computing operations, reducing bus resource occupation while maintaining execution efficiency.
Solution Approach 2:
An accelerator is introduced as an intermediary component between the processor and memory. The accelerator handles data computing operations independently, acting as a mediator that performs computations without requiring the processor to constantly access memory through the bus, thereby reducing bus resource occupation.
2Productivity
If the processor performs all data computing, then computing accuracy is maintained, but processor overhead increases and execution efficiency decreases
Solution Approach 1:
The computing system is segmented into a processor for control operations and an accelerator for data computing operations. This division reduces processor overhead by transferring data computing responsibilities to the accelerator, while the processor maintains overall system control and accuracy.
Solution Approach 2:
The accelerator is designed to perform data computing operations autonomously without requiring continuous processor intervention. It can independently access memory, perform computations, and manage its own operations, thereby reducing processor overhead while maintaining computing accuracy through proper design.
Data Source
Figure 1~2
Figure 3~4
Figure 5~6
AI summary
The present disclosure provides a data computing system. The data computing system comprises: a memory, a processor and an accelerator. The memory is configured to store data to be computed, the processor is communicatively coupled to the memory and configured to write the data to the memory, and the accelerator is communicatively coupled to the memory and the processor and configured to receive control information from the processor and to access the memory according to the control information, to implement a computing process that produces a computed result, and to write the computed result back to the memory. The received control information comprises a start address for the data to be computed, a computing type, and a write-back address for the computed result, and the computing process is implemented by the accelerator independently from the processor.