双曲型方程式のGPUによる高速計算(<特集>GPGPUコンピューティングの数理)

DOI Web Site 参考文献6件

書誌事項

タイトル別名
  • GPU Acceleration for Computaion of Hyperbolic Equations(<Special Topics>GPGPU Computing)
  • 双曲型方程式のGPUによる高速計算
  • ソウキョクガタ ホウテイシキ ノ GPU ニ ヨル コウソク ケイサン

この論文をさがす

抄録

In order to solve hyperbolic equations such as electromagnetic wave equation, GPU(Graphics Processing Unit) is used as an accelerator. The characteristics of the physical phenomena modeled by hyperbolic equations make it easy to apply the GPU to highly-parallel thread computing, since explicit scheme is available for them. We demonstrate a compressible flow solving Euler equation with a higher-order numerical scheme IDO-CF and we achieved 54 GFlops on NVIDIA GeForce GTX 285. The multi-GPU computing for lattice Boltzmann method is carried out on the Tesla GPU of TSUBAME grid cluster of Tokyo Tech and the strong scaling is investigated up to 100GPUs. The shallow water equation for the real-time tsunami simulation is also solved on the multi-GPU and the overlapping technique between the communication and computation to hide the GPU-to-GPU communication time. The Tesla 10GPUs achieved the performance comparable with 1000 TBUBAME Opteron CPUs.

収録刊行物

  • 応用数理

    応用数理 20 (2), 94-106, 2010

    一般社団法人 日本応用数理学会

参考文献 (6)*注記

もっと見る

詳細情報 詳細情報について

問題の指摘

ページトップへ