GPU sharing control method, GPU sharing control device and storage medium

Abstract

The invention provides a GPU sharing control method, a GPU sharing control device and a storage medium. The GPU sharing control method comprises the steps that in response to starting of all deep learning tasks, all the deep learning tasks are divided into main tasks or secondary tasks, and the number of the main tasks is not larger than one. And determining operation stage information of each deep learning task, wherein the operation stage information is divided into an operation stage occupying the GPU and an operation stage not occupying the GPU. And judging whether the main task is in a GPU operation occupying stage or not. And when the main task is in the stage of occupying GPU operation, controlling the GPU to operate the main task. And when the main task is not in the GPU operation occupying stage, controlling the GPU to operate the secondary task. Through the method provided by the invention, it can be preferentially guaranteed that the GPU can be controlled to prioritize the main task when the main task is in the GPU operation occupying stage according to the operation stage information of each deep learning task, and thus the Qos of the main task is guaranteed. Meanwhile, when the main task is not in the GPU operation occupying stage, GPU computing power can be fully utilized, and the utilization rate of the GPU is increased.

Description

technical field [0001] The present invention relates to the technical field of computer control, in particular to a GPU sharing control method, a GPU sharing control device and a storage medium. Background technique [0002] In the era of artificial intelligence and big data, with the increasing amount of data and the development of deep learning algorithms, the demand for computing power is also increasing. The combination of deep learning tasks and GPUs is used in various industries. However, in the process of executing deep learning tasks, the demand for GPU computing power fluctuates, and according to the execution logic of the kernel (kernel) in the underlying context of the GPU, a single deep learning task often cannot use the computing power of the GPU at all times, resulting in Waste of GPU computing power. [0003] In related technologies, multiple deep learning tasks can be scheduled to the same GPU, and the computing power of the GPU can be shared to control mult...

AI Technical Summary

Problems solved by technology

[0004] Therefore, the technical problem to be solved by the present invention is to overcome the defects in the prior art that the utilization rate of GPU computing power is low when GPU sharing is performed, resulting in waste of computing power, thereby providing a GPU sharing control method, a GPU sharing control device and storage medium

Images