A kind of onboard graphics processor control method and device

Abstract

The invention discloses a method and device for controlling an onboard graphics processor, comprising: accessing a substrate controller to obtain configuration information, configuring a high-speed serial computer expansion bus switching chip, a clock multiplexer, and a clock buffer, using a power supply signal to communicate with the board The on-board voltage regulator communicates to perform power-on and timing control, so that the high-speed serial computer expansion bus switch chip connects a plurality of onboard graphics processors to the high-speed serial computer expansion bus and provides clock signals according to the configuration information. The invention can provide PCIE interface and clock signal according to multiple different GPUs or different working states of GPU, control the onboard graphics processor to work normally according to its performance requirements, and improve the processing capability of the onboard graphics processor.

Description

technical field [0001] The present invention relates to the field of computers, and more particularly, to a method and device for controlling an onboard graphics processor. Background technique [0002] With the continuous development of Internet technology and the increasing demand for parallel computing, GPU (graphics processing unit) can effectively release computing pressure with its excellent graphics processing capabilities and high-performance computing capabilities, and significantly improve the computing processing efficiency and competitiveness of products. GPU servers are gradually emerging and showing an intensifying trend. In order to give full play to the computing performance of the GPU, the GPU installation method in the GPU server has gradually changed from the traditional GPU plug-in card form to the onboard GPU form. Unlike the plug-in GPU, the onboard GPU needs to provide multiple sets of PCIE signals for the GPU board to realize the data interaction bet...

AI Technical Summary

Problems solved by technology

Different from the plug-in GPU, the onboard GPU needs to provide multiple sets of PCIE signals for the GPU board to realize the data interaction between the CPU (central processing unit) and each GPU. Bus) ports are limited, and the number of concurrency is small, which cannot improve the computing power of the GPU; on the other hand, different GPUs may have different operating frequencies, or the same GPU may have different operating frequencies in different working states, and the current Some technologies cannot arbitrarily switch the clock signal according to the operating frequency requirements

[0003] Aiming at the problems of limited number of GPU ports and difficult switching of clock signals in the prior art, there is currently no effective solution

Images