A method and system for realizing CPU and GPU load balancing

Abstract

The present application discloses a method and system for realizing CPU and GPU load balancing, including: obtaining computing device information, physical memory and other system resource related information currently configured in the system; according to the obtained computing device information, physical memory and other system resources Relevant information, determine the type and quantity of the computing equipment to be started; according to the type and quantity of the computing equipment to be started, create the control thread of the corresponding computing equipment; to realize independent and parallel operation of each computing equipment; computing equipment includes: central processing unit ( CPU), and / or graphics processing unit (GPU). The present invention records and manages resource-related information such as computing device information and physical memory configured by the system through the data structure of device attribute information, and creates corresponding control threads after determining the type and quantity of the computing device to be started, realizing the cooperation between CPU and GPU Parallel computing; in addition, the task load is distributed through the acceleration ratio of each computing device relative to the CPU, and the load balancing of the CPU and GPU is realized.

Description

technical field [0001] The present application relates to the field of computers, in particular to a method and system for realizing load balancing between a central processing unit (CPU) and a graphics processing unit (GPU). Background technique [0002] Driven by new technologies, hardware systems are undergoing rapid evolution, showing better performance and lower prices. CPU and GPU play an important role in the evolution of hardware systems. CPU is constantly developing from single-core to multi-core to many-core, and with the popularity of multi-core architecture processors, the parallel processing method of multi-thread application software has been paid attention to. In addition to traditional applications (graphics display, mostly used in games), GPU is increasingly used in mathematical calculations due to its super floating-point computing capabilities, and has gradually become the mainstream of mathematical calculations. High-level languages ​​and development to...

AI Technical Summary

Problems solved by technology

Among them, dynamic load balancing means that each computing device adopts the strategy of "capable people do more work" according to their actual computing capabilities, real-time, dynamic, and autonomous task load distribution. Dynamic load balancing is flexible and efficient, but it has strict requirements on algorithms. Computing tasks are required to be completely independent of each other, so that they can be processed in parallel; static load balancing means that each computing device allocates computing tasks according to a preset and fixed task load distribution ratio. Static load balancing is simple to implement, but Lack of flexibility and adaptability

These two load balancing methods have their own limitations and are not suitable for all occasions. For example, in high-performance parallel computing occasions where the input and preprocessing of basic computing tasks overlap and are redundant, complete dynamic load balancing will lead to a large amount of redundancy. Input and preprocessing lead to excessive overhead, resulting in poor overall efficiency; while complete static load balancing can minimize the redundancy of input and preprocessing, but it cannot be dynamically adjusted in real time according to actual operating conditions Calculation task allocation ratio, serious lack of flexibility and adaptability, use effect is not good

[0005] In summary, the algorithmic limitations of complete dynamic load balancing and the lack of flexibility and adaptability of static load balancing affect the performance of CPU and GPU collaborative parallel computing.

Images