Initialising ...
Initialising ...
Initialising ...
Initialising ...
Initialising ...
Initialising ...
Initialising ...
Idomura, Yasuhiro
no journal, ,
In performing extreme scale CFD simulations on many core platforms with low power consumption such as Oakforest-PACS, we need new computing technologies for accelerating computation on many core processors, and avoiding communications and data I/O, which become bottlenecks with accelerated computation. In order to resolve these issues, we have developed many core optimization techniques, communication latency hiding techniques, communication avoiding algorithms, and In-Situ visualization systems on Oakforest-PACS. In this talk, we present applications of these techniques to five dimensional plasma CFD codes and three dimensional multi-phase thermal hydraulic CFD codes.
Kawamura, Takuma; Idomura, Yasuhiro; Onodera, Naoyuki
no journal, ,
In-situ PBVR enables an interactive visualization of batch processing simulations on CPU supercomputer. In this research, we developed novel particle generation method of In-situ PBVR for AMR mesh, and ported the in-situ PBVR framework into CPU of GPU platform for task parallel visualization. We applied the proposed method to thermal convection simulation on a GPU cluster, and confirmed that the proposed method has the same image quality as a visualization software ParaView and is 30 times faster than it.
Yano, Midori; Kawamura, Takuma; Hasegawa, Yuta; Idomura, Yasuhiro
no journal, ,
Visualizing a large amount of ensemble data and time series data generated from extreme scale simulations is difficult because of limitations on the performance of data I/O, data transfer, and visualization. In-Situ visualization techniques that visualize simulations at runtime are promising solutions for this problem. This study proposes an in-situ statistical analysis method, applies it to time-series data of a real-time plume dispersion simulation, and analyzes winds and concentration statistically using a scatter plot and glyph in a target 3D space. Users can map statistical data to glyph design and visualize statistical properties of the simulation interactively.