An accurate model to predict the performance of graphical processors using data mining and regression theory. (March 2021)

• Record Type: Journal Article
• Authors: Shafiabadi, Mohammadhossein; Pedram, Hossein; Reshadi, Midia; Reza, Akram
• Is Part Of: Computers & electrical engineering. Volume 90(2021)
• Publication Date: 2021-03
• Subjects: Data mining -- Graphical processors unit -- Nonlinear regression model -- Instruction per second -- Prediction -- Statistical analysis -- Residual; Computer engineering -- Periodicals; Electrical engineering -- Periodicals; Electrical engineering -- Data processing -- Periodicals; Ordinateurs -- Conception et construction -- Périodiques; Électrotechnique -- Périodiques; Électrotechnique -- Informatique -- Périodiques; Computer engineering; Electrical engineering; Electrical engineering -- Data processing; Periodicals; Electronic journals; 621.302854
• DOI: 10.1016/j.compeleceng.2020.106965
• Languages: English
• ISSNs: 0045-7906

Abstract

Highlight: Using non-linear regression model to predict performance of graphic processor unit. Using data mining approach to enhancement performance predicted in graphic processor unit. Using application parameter and hardware parameter in generate model to predict accurately performance of graphic processor unit. Using statistical analysis to verify the accuracy if the model. Abstract: Nowadays the use of graphical processors in fast and accurate scientific calculations has increased. The heterogeneous design space that is conducted by the processors could provide important assistance for the designers to achieve suitable accuracy, although preparing a proper model to predict the performance of these processors is very difficult. In this paper, first, the relationship between independent parameters in design space and a dependent parameter that is processor performance indicated by instructions per cycle was investigated. The design space was made smaller and then by using statistical inference for regression, an accurate model was proposed to predict the performance of graphical processors. The proposed model was evaluated by using AMD Southern Island and SDK 2.5 benchmark applications. Based on the extensive result, the proposed model could successfully predict the performance with an error rate of 6%. In addition, both analyses evaluated the accuracy of the model by approximately 95%. Graphical abstract: Image, graphical abstract