A study of the nighttime heat island and the acceleration of LES on the multi-GPU cluster

  • Kazunari TAKAHASHI, 2014: A study of the nighttime heat island and the acceleration of LES on the multi-GPU cluster .


The formation of the heat island there are a lot of factors. Factor is two roughly classified into of thermal effects and mechanical effects, discussion respectively have been made. In particular, the formation of the "nighttime heat island", in addition to the thermal factors such as the building heat storage release, have affected from the mechanical turbulence generated from the building. However, heat island is the complexity of the size and components of the spatial scale, observation is difficult, verification by numerical experiments is required. In previous heat island numerical simulations, because of incorporating the effects of city as a parameter into the model, also of spatial resolution is low, the formation of nighttime heat island detailed process is not known. In this study, we performed numerical experiments using the Large Eddy Simulation (LES) about the contribution of mechanical effects in the nighttime heat island formation by ground inversion layer breaking. As a result, in the urban area, the heat is likely to pooling performed by wind drag effects. Also, the mechanical turbulence from buildings are mixing the air and increasing the temperature. In addition, we studied about the acceleration of LES using Graphics Processing Unit(GPU). because of the heat island numerical experiment needs a lot of computing resources. As a result, the performance of each function, achieved a speed-up of 1.6 ~ 55.2 times in 4GPU versus 4CPU-core. In 4GPU versus 16CPU-core achieved 0.4 ~ 25.1 times faster. The performance per time step is 1.8 times in 4GPU versus 4CPU-core, achieved a speed of 4.3 times in 4GPU versus 16CPU-core.