在日中国人CFD研究会講演会

第三十九回講演会　　講演１
• 講師：　　　武　珑　博士　（エクサ・ジャパン株式会社　プロダクトエンジニア）
• 講演題目：　An overview of the theory and applications of Exa’s flagship product　“PowerFLOW”
• 開催日：　　2010年02月06日（土曜日）午後14:00～16:00
• 開催地：　　理化学研究所・研究交流棟3F会議室(W319,W321)
• 写真集
• 理化学研究所へのアクセス
• 会場の研究交流棟（B-7の36番建物）
• 講演概要：

  Exa PowerFLOW is an unique solution for computational fluid dynamics (CFD) problems, suitable for a wide range of applications, whether simulating transient or steady-state flows. PowerFLOW uses an enhanced lattice-Boltzmann method, in which distribution of particles exist at discrete locations in space and move in discrete directions, speeds and intervals of time. PowerFLOW simulates fluid flow design problems in such industries as automotive and other ground transportation, aerospace, petroleum, building design/AEC, HVAC, and electronics.

  Features, Benefits and Highlights:
   Fluids Solver
  　　 Fast, time dependent solver for unsteady fluid mechanics
  　　 Disruptive technology validated by industry
  　　 Patented extended Lattice Boltzmann solver
  　　 More fundamental and accurate approach than Navier-Stokes PDE solvers
   Other benefits:
  　　 No significant artificial dissipation
  　　 Excellent scalability up to hundreds of processors
  　　 Robust & physical realization of boundary conditions
   Grid Generation
  　　 Immersive grid technology
  　　 Fully automatic fluid grid generation
  　　 Captures all details of complex geometry
  　　 No limit on geometric complexity – no need to simplify
  　　 Unlike most conventional CFD, cell quality is not a concern
  

• 講師略歴
  1999年 4月　中国西安交通大学能源与動力工程学院　学士
  2002年 4月　中国西安交通大学能源与動力工程学院　修士
  2008年 3月　日本神戸大学機械工学研究科　工学博士
  2008年 4月　エクサ・ジャパン株式会社　プロダクトエンジニア
  

  
  

第三十九回講演会　　講演２
• 講師：　　　王　嫻　博士（東京工業大学理工学研究科　JST-CREST研究員）
• 講演題目：　CUDA programming and large-scaled computation by Lattice Boltzmann method on single GPU and multi-node GPU cluster
• 開催日：　　2010年02月06日（土曜日）午後16:00～18:00
• 開催地：　　理化学研究所・研究交流棟3F会議室(W319,W321)
• 写真集
• 理化学研究所へのアクセス
• 会場の研究交流棟（B-7の36番建物）
• 講演概要：

  A Graphical Processing Unit (GPU) is specifically designed to be extremely fast at processing large graphics data sets (e.g., polygons and pixels) for rendering tasks. Recently, the use of GPU to accelerate non-graphics computations has drawn much attention. Due to its high performance of floating-point arithmetic operation, wide memory bandwidth and better programmability, GPU has been further used for general purpose with the release of CUDA (Compute Unified Device Architecture) by NVIDIA. That is GPGPU (General-Purpose Graphics Processing Unit). A various applications have already achieved high performance on GPU. The mesh-based CFD applications for engineering purpose is one of the most important application of GPGPU. As an example, Lattice Boltamann method is successfully executed on GPU. The speed-up of LBM solver on single GPU is more than 80 over CPU, which is much higher than that of NSE based method. The performance is about 50 GFLOPS. On multi-node GPU cluster (TSUBAME Tokyo titech), 1800 GFLOPS performance is obtained for 3D-LBM solver by 96 GPUs. However, lots of time is consumed on data transfer between CPU-CPU and CPU-GPU for multi-GPU computation, especially for 1D-decomposition mode. To solve this problem, multi-decomposition mode and asynchronous memory operation technique can be used. Finally, a large-scaled computation on the flow around a sphere is carried out on with Re=13000, mesh= 2000x1000x1000 and 100 GPUs. As a result, 9 hours is used for 100,000 LBM loops. The ratio between computational time and data communicational time is about 1.6.

• 講師略歴
  1999年　中国西安交通大学能源与動力工程学院　卒業
  2002年　中国西安交通大学能源与動力工程学院　修士修了
  2005年　日本九州大学機能物質研究所　博士修了
  2005年　日本岡山県産業振興財団　研究員　
  2008年　日本東京工業大学理工学研究科　JST-CREST研究員