high performance computing on graphics processing units: hgpu.org

Today, there is no one who disagrees on how important data is in every industry especially in enterprise market. More recently, the key point that decides the survival of a business is the management of their big data, which is defined by the 3V’s: Volume, Velocity, and Variety [1]. While the rate of data generation increases exponentially, processing that data with the limited resources can be a burden to the both business managers and IT managers. Therefore many researchers have already studied new systems which can serve as an alternative resource to calculate and process data. Parallel hardware, such as a general-purpose GPU (GPGPU), is one of the most well-known alternative. With them, it is possible to process various applications, including data-intensive applications, quickly [2]. OpenCL, in collaborated with several GPU vendors and software organizations, has been launched by the Khronos group as the first open standard platform for the programming of both the GPUs and CPUs [3]. It makes the binary codes execute on various heterogeneous processing units such as CPUs, GPUs and FPGAs simultaneously. It also supports small clients like mobile GPUs for the mobile world. This paper proposes the method to optimize a machine learning algorithm with the heterogeneous platform which uses both the CPUs and GPUs using OpenCL. Through the experiment, we show that our method can reduce the execution time of the k-means nearest clustering algorithm, which is one of the most common algorithms in the machine learning industry, up to 40%. The more data we use in our system, the faster our results are when compared to the experiment in the multi-core system.