Heterogeneous parallel computing device and computer node internetwork thereof

Abstract

The invention provides a heterogeneous parallel computing device and a computer node internetwork thereof. Real-time signal processing and parallel computing for radar, communication and artificial intelligent can be performed; the heterogeneous parallel computing device comprises one central control node, multiple GPU (graphics processing unit) computer nodes and one computer node internetwork; the computer node internetwork adopts a star network, the central control node can perform point-to-point communication with all the GPU computer nodes, and the heterogeneous parallel computing device is used for transmitting real-time control information and performing real-time parallel computing on tasks and data and applicable to occasions that executed computation has sudden and parallel characteristics; the computer node internetwork adopts an annular network, the central control node assigns tasks to the GPU nodes step by step, processed data can be transferred through the annular network step by step with no need of the central control node, data throughout of the central control node is alleviated, and the computer node interwork is applicable to large-scale pipelined computation processing occasions.

Description

technical field [0001] The invention relates to the technical field of high-performance computing applied to radar, communication, and artificial intelligence, and specifically relates to a heterogeneous parallel computing device and its computing node interconnection network. Background technique [0002] In the high-performance computing fields of radar, communication, and artificial intelligence, the requirements for computing data throughput, computing power per unit time, and real-time processing are getting higher and higher. In these fields, traditional processors mainly include MCU, CPU, DSP, FPGA and so on. From the perspective of system development and software programming, the development difficulty of the above-mentioned devices generally has a trend from easy to difficult. However, from the perspective of real-time performance of large-scale computing, MCU and CPU have the lowest processing real-time performance, DSP is in the middle, and FPGA is the highest. ...

AI Technical Summary

Problems solved by technology

[0004] 1) In occasions that require strong real-time computing, such as radar imaging and high-speed communication, there is a lack of means for strict timing control;

[0005] 2) Multi-GPU systems often use complex network connections, which require more system overhead and further reduce the computing power of the system;

[0006] 3) It is suitable for occasions where PCs and servers are used, but cannot meet the needs of a large number of embedded devices

Images