Comparison of Intel Core-i7, Intel Xeon, Intel Xeon Phi and IBM Power 8 processors on the example of the task of restoring initial data

A.Yu Gorchakov, V.U. Malkova

Abstract


In this paper, a comparative analysis of four types of processors is performed using the example of the problem of restoring the initial data for the transport equation. The problem is solved by the Levenberg-Marquardt method, which is decomposed into four subtasks - calculation of the vector function, calculation of the matrix of the first derivatives of the vector-function (the Jacobi matrix), matrix multiplication and the solution of the system of linear equations. The calculation of the Jacobi matrix is performed using the methodology of fast automatic differentiation, using the application software package Adept version 1.1. To speed up the calculations, we used directive multithreaded programming with the dynamic distribution of computations between threads and SIMD (Single Instruction Multiple Data) technology, the principle of computer computation, which allows to provide parallelism at the data level. The aforementioned technologies allowed to fully exploit the features of the architecture of modern Intel processors, such as multi-core / multithreading, the expansion of the command system of microprocessors Intel / AMD - Advanced Vector Extensions (AVX / AVX2), processing data in floating point format in groups of 256 bits, and FMA (Fused Multiply-Add) - a technology designed to perform a combined multiply-add operation. In comparison, the processors Intel Core i7 4770 (Haswell), Intel Xeon E5-2683V4 (Broadwell), Intel Xeon Phi coprocessor SE10 / 7120 and IBM Power 8 are given. Recommendations are given for choosing the type of processor depending on the task being solved.

Full Text:

PDF (Russian)

References


Hogan R. J. Fast reverse-mode automatic differentiation using expression templates in C++ //ACM Transactions on Mathematical Software (TOMS). – 2014. – T. 40. – #. 4. – S. 26.

Anderson E. et al. LAPACK Users' guide. – Society for Industrial and Applied Mathematics, 1999

Federal'nyj issledovatel'skij centr Informatika i upravlenie RAN [Jelektronnyj resurs]: sajt. – Moskva: FIC IU RAN. – URL: http://frccsc.ru

Evtushenko Ju. G., Zubov V. I. Ob obobshhennoj metodologii bystrogo avtomaticheskogo differencirovanija. Zhurnal vychislitel'noj matematiki i matematicheskoj fiziki, 2016, vol. 56 (11), pp. 1847-1862.

Abgarjan K. K., Posypkin M. A. Programmnyj kompleks dlja reshenija zadach parametricheskoj identifikacii potencialov mezhatomnogo vzaimodejstvija //International Journal of Open Information Technologies. – 2014. – T. 2. – #. 10.

Gill F., Mjurrej U., Rajt M. Prakticheskaja optimizacija. – 1985.

Evtushenko Ju. G. Optimizacija i bystroe avtomaticheskoe differencirovanie //M.: Nauchnoe izdanie VC RAN. – 2013.

Gorchakov A. Ju. Ispol'zovanie vozmozhnostej OpenMP dlja uskorenija raschetov v pakete prikladnogo programmnogo obespechenija Adept // CEUR Workshop Proceedings (CEUR-WS.org): Selected Papers of the II International Scientific Conference Convergent Cognitive Information Technologies (Convergent 2017), Moscow, Russia, November 24-26, 2017. — 2017. — P. 217-223. DOI 10.25559/SITITO.2017.6.630


Refbacks

  • There are currently no refbacks.


Abava  Кибербезопасность IT Congress 2024

ISSN: 2307-8162