A multi-gpu high-performance processing system based on the openvpx platform

Abstract

The invention discloses a multi-GPU high-performance processing system based on an OpenVPX platform. The system comprises two dual-graphics processor GPU processing boards, a central processing unit CPU processing board and a 6-slot VPX backboard. The dual GPU processing board has two MXM modules equipped with GPU and a CPLD control chip. The 6-slot VPX backplane includes seven VPX connectors fromP0 to P6, of which P2 and P5 are configured as VPX connectors supporting the high-speed serial computer expansion bus standard PCIe communication standard, and P2 and P5 each have two 8x PCIe interfaces. CPU CPU processing board with a CPU mounted on the ComE module and a 64-channel PCIe switch chip; the ComE module is connected to the PCIe switch chip via 16x PCIE. A total of four MXM modules onthe two GPU processing boards are respectively connected to a total of four 8x PCIe interfaces on P2 and P5 on the VPX backplane through 8x PCIe; a total of four 8x PCIe interfaces for P2 and P5 on the 6-slot VPX backplane are connected to the PCIe switch chip.

Description

technical field [0001] The invention relates to the technical field of real-time signal processing, in particular to a multi-GPU high-performance processing system based on the OpenVPX platform. Background technique [0002] With the vigorous development of science and technology and the advent of the high-speed information age, the increasing amount of dense signal data puts forward higher requirements on the processing capacity of the processing system, and greater processing capacity is required per unit volume and unit power consumption. The processing performance of single-processor architecture computing platforms is often difficult to meet the needs of practical applications. Increasing the number of processors in the processing system can better meet the performance requirements of large-scale computing. Therefore, multi-processor computing platforms have become an important part of high-performance computing a new solution. [0003] For the application field of sig...

AI Technical Summary

Problems solved by technology

[0004] At present, most of the system designs are completed with a single GPU, and most of the systems have poor configurability, and various requirements cannot be realized under the same architecture. The difficulty, cost, and cycle of development are difficult to meet the increasing quantity requirements of existing ground systems. , the research on multi-GPU implementation scheme still has great progress and room for improvement

Images