Initialising ...
Initialising ...
Initialising ...
Initialising ...
Initialising ...
Initialising ...
Initialising ...
Utsuno, Yutaka
no journal, ,
no abstracts in English
Idomura, Yasuhiro; Maeyama, Shinya*; Nakata, Motoki*; Nunami, Masanori*; Ishizawa, Akihiro*; Watanabe, Tomohiko*
no journal, ,
A novel parallel optimization technique for extreme scale CFD simulations is developed on the K-computer, and strong scaling of finite difference and spectral fusion plasma turbulence codes is improved toward million-core regimes. The optimization technique consists of a multi-dimensional and multi-layer domain decomposition, optimized process mapping on a three dimensional (3D) torus network, and overlap of computations and communications using communication threads. The improved strong scaling dramatically expanded capabilities of the fusion plasma turbulence codes both in problem sizes and time-scales, and enabled us to study critical issues in ITER such as the plasma size scaling of ion turbulent heat transport, and the electron heat transport induced by multi-scale electron turbulence.
Hirota, Yusuke*; Yamada, Susumu; Imamura, Toshiyuki*; Sasa, Narimasa; Machida, Masahiko
no journal, ,
The numerical solution of the eigenvalue problems generally becomes less accurate as the matrix dimension increases due to rounding errors. To overcome the accuracy problem, in 2012, we started to develop a quadruple precision version of eigensolver libraries for a standard eigenproblem and a generalized eigenproblem: QPEigenK and QPEigenG, respectively. In this study, we evaluate the performance and the accuracy of QPEigenK and QPEigenG on the K computer. The results show that the solvers of the libraries compute more accurate eigenpairs of large problems in reasonable time and show high scalability.