Результаты исследований: Глава в книге, отчете, сборнике статей › Материалы конференции › Рецензирование
Результаты исследований: Глава в книге, отчете, сборнике статей › Материалы конференции › Рецензирование
}
TY - GEN
T1 - The Effect of Data Structuring on the Parallel Efficiency of the HydroBox3D Relativistic Code
AU - Chernykh, Igor
AU - Misilov, Vladimir
AU - Akimova, Elena
AU - Kulikov, Igor
N1 - Computations were performed on the NKS-1P supercomputer at the Siberian Supercomputer Center, Institute of Computational Mathematics and Mathematical Geophysics SB RAS, Novosibirsk, Russia. This work was supported by the Russian Science Foundation (project 23-11-00014) https://rscf.ru/project/23-11-00014/.
PY - 2024/1/5
Y1 - 2024/1/5
N2 - The hydrodynamic approach to modeling astrophysics problems has several disadvantages in terms of the implementation of a parallel computing code. One of the main drawbacks is the low arithmetic intensity of the methods that implement the computational problem. This peculiarity produces the performance limitation associated with the performance limitations of the DRAM memory of high-performance computing systems. One of the solutions to this problem is data structuring based on the characteristics of processors and memory of a computer system on which supercomputer simulation is to be carried out. In this work, the authors use the specialized Intel SDLT library, which allows you to organize data in a special way that can help the compiler to vectorize a computational code for Intel server processors. The use of this library made it possible to speed up the computational code by fifty times, and for the first time bring the performance of some code functions to the performance limits of server processors on vector FMA instructions.
AB - The hydrodynamic approach to modeling astrophysics problems has several disadvantages in terms of the implementation of a parallel computing code. One of the main drawbacks is the low arithmetic intensity of the methods that implement the computational problem. This peculiarity produces the performance limitation associated with the performance limitations of the DRAM memory of high-performance computing systems. One of the solutions to this problem is data structuring based on the characteristics of processors and memory of a computer system on which supercomputer simulation is to be carried out. In this work, the authors use the specialized Intel SDLT library, which allows you to organize data in a special way that can help the compiler to vectorize a computational code for Intel server processors. The use of this library made it possible to speed up the computational code by fifty times, and for the first time bring the performance of some code functions to the performance limits of server processors on vector FMA instructions.
UR - http://www.scopus.com/inward/record.url?partnerID=8YFLogxK&scp=85182597934
U2 - 10.1007/978-3-031-49432-1_21
DO - 10.1007/978-3-031-49432-1_21
M3 - Conference contribution
SN - 978-303149431-4
VL - 14388 LNCS
T3 - Supercomputing
SP - 271
EP - 284
BT - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
A2 - Voevodin, V.
A2 - Sobolev, S.
A2 - Yakobovskiy, M.
A2 - Shagaliev, R.
PB - Springer
ER -
ID: 51655911