Intelligent acceleration card and data processing method based on intelligent acceleration card

Abstract

The embodiment of the invention discloses an intelligent acceleration card and a data processing method based on the intelligent acceleration card. A computing resource and an operating system are configured on the intelligent acceleration card, the operating system is at least divided into a kernel and a user space which are respectively used for controlling the computing resource and running an application program, the computing resource provided by physical hardware of the intelligent acceleration card is mapped into a plurality of containers in the user space, the calling of the computing resources by the application programs can be realized in the container, so that the operation among the application programs is isolated, and mutual interference is avoided. In the process of calling the computing resources by the application program, after the application program of the user space sends out the data processing request, the kernel hijacks the data processing request and determines the resource allocation information of the computing resources matched with the data processing request, so that resource scheduling and management can be realized, the utilization rate of the resources is improved, and the user space can call the corresponding computing resources to execute the data processing operation corresponding to the data processing request based on the resource allocation information.

Description

technical field [0001] The present application relates to the technical field of data processing, in particular to a data processing method based on an intelligent accelerator card, a data processing device based on an intelligent accelerator card, an operating system, and an intelligent accelerator card. Background technique [0002] As an accelerator chip, smart accelerator cards are widely used in applications that require high computing resources, such as machine learning, especially deep learning. [0003] When enterprises, schools, hospitals or management departments invest in the field of artificial intelligence, they usually need to purchase a large number of smart accelerator cards to obtain sufficient computing resources. In practical applications, it is found that the use of computing resources of the smart accelerator card has the following problems: [0004] 1) The purchase cost of the smart accelerator card is relatively high, and the computing resources are v...

AI Technical Summary

Problems solved by technology

[0004] 1) The purchase cost of the smart accelerator card is relatively high, and the computing resources are very expensive. With the increasing computing demand, the computing resources are always not enough

At the same time, due to the limitation of technological development, in actual use, it is difficult to share computing resources, which leads to the coexistence of waste and shortage

[0005] 2) The demand for usage scenarios is high. When doing machine learning training, a large amount of computing resources are required and it needs to be used for a long time. When doing online reasoning, it needs to respond in a timely manner, especially when there are many concurrent users in scenarios such as development, training, and course training. , sufficient computing resources are needed to meet the timely response of a large number of users

[0006] 3) The construction and configuration of the computing environment for machine learning is very complicated, and usually data engineers are not good at it. How to optimize the use of computing resources is a relatively big challenge

Images