Embedded intelligent computer system for object data processing

Abstract

The present invention proposes an embedded intelligent computer system oriented to object data processing, including: an IO interconnection system, which is used to send object data from multi-source sensors to the artificial intelligence processor subsystem through a high-speed data channel, and The tag category of the object data is sent to the embedded ARM processor subsystem; the embedded ARM processor subsystem is used to retrieve the artificial intelligence neural network model corresponding to the tag category in the memory, and the artificial intelligence neural network model The network model is loaded into the artificial intelligence processor subsystem; the artificial intelligence processor subsystem is used to input the object data into the artificial intelligence neural network model to complete the artificial intelligence processing of the object data and obtain intelligent data process result. Therefore, this application can not only complete the natural environment data collection and calculation tasks, but also complete the human factors environment data collection and calculation tasks.

Description

technical field [0001] The invention belongs to the technical field of embedded systems in computer architecture, and in particular relates to an embedded intelligent computer system oriented to object-end data processing. Background technique [0002] Object computing mainly refers to the general designation of various sensor devices, terminal computing devices, user terminal computing devices and their internal computing tasks for natural environment data collection and computing tasks and human environment data collection and computing tasks, including sound, light, electricity, Magnetic, thermal, energy and other sensors and their corresponding computing modules are used in various computing environments such as mobile phones, personal computers, industrial robots, vehicles, ships, airborne, and spaceborne. The most typical representatives are visible light cameras, infrared Cameras, navigation radars, millimeter wave radars, ultrasonic radars, target detection radars, o...

AI Technical Summary

Problems solved by technology

[0005] (2) Cannot meet the requirements of multi-task concurrent intelligent computing

The overall calculation beat is executed serially, and parallel loading and calculation of multiple neural network models cannot be realized

[0006] (3) Unable to meet the power consumption requirements of intelligent computing tasks (including neural network model training and reasoning involved in deep learning and reinforcement learning)

Usually GPU adopts multi-core parallel computing architecture, its data throughput far exceeds that of CPU, but its power consumption is also very large

For example, the data throughput of the NVDIA Tesla V100 GPU is 47 times that of the server CPU, but its power consumption is also very large, exceeding 200W, which is only suitable for cloud training and cloud reasoning, not for object-side reasoning

Images