vGPUPCD: Design and Implementation of Virtual GPU on Private Cloud Desktop

Abstract

This paper proposes to design and implement GPU virtualization on the Private Cloud Desktop merged into the Teaching and Practical Platform of Zhijiang College of Zhejiang University of Technology to fill in the blanks to create and apply rending 3D applications. The analysis of the performance of the vGPUPCD has been also studied in this paper and some properties has to be tested on the experiments: 1) The video memory size of vGPU has been interaction with the memory size and the amounts of CPU cores and the minimum resources requirement has to be evaluated by experiments that the memory size is the memory size of system demand and add to the twice of the GPU video memory size and the 1G video memory of vGPU is equipped with 2 virtual CPU cores and so on. 2) The GPU memory usage of vGPUPCD has consumed 10% more than the local PCs if the same performance are reflected when opening or rending the 3D applications because vGPU has occupied no more than 10% of the GPU memory usage through the cloud desktop is powered on.