Computer prediction of protein-protein interaction network using MEGADOCK

In order to drastically shorten the runtime of the weather-prediction code ASUCA , developed by the JMA (Japan Meteorological Agency) for the purpose of the next-generation weather forecasting service, the entire parts of the huge code were rewritten for GPU computing from scratch. By introducing many optimization techniques and several new algorithms, very high performance of 145 TFlops has been achieved with 3990 GPUs on TSUBAME 2.0 Supercomputer. It is quite meaningful to show that the GPU supercomputing is really available for one of the major applications in the HPC field. Weather forecasting is an indispensable part in our daily lives and business activities, needless to say for natural disaster preventions. The atmosphere has a very thin thickness compared with the Earth diameter. In the previous atmosphere code, the force balance between the gravity and the pressure gradient in the vertical direction was used to produce a hydrostatic model. Recently it is widely recognized that the vertical dynamical processes of the water vapor should be taken into consideration in cloud formations. A three-dimensional non-hydrostatic model describing up-and-down movement of air has been developed in weather research. For weather simulations, the initial data is produced by assimilating many kinds of observed data and simulation results based on the four-dimensional variational principle. Since the weather phenomena are chaotic, the predictability period is less than several days for one set of initial data hence the jobs run sequentially updating the initial data. In recent years, it is highly demanded to forecast detailed weathers such as unexpected local heavy rain, and high resolution non-hydrostatic models are desired to be carried on fine-grained grids. A computational heavy load is required to run the high-resolution weather models. As a reference the WRF [1] has been scored at 50 TFlops on the current fastest supercomputer in the world [2]. WRF is a next-generation atmosphere simulation model (Weather Research and Forecasting), a world standard code developed at the Numerical weather models consist of a dynamical core and physical processes. In the dynamical core, forecast variables such as winds, atmospheric pressure and humidity are calculated by solving fluid dynamics equations. The physical processes strongly depend on parametrizations related to such microphysics as condensation of water vapor, cloud physics, and rain. In the computation of the dynamics core, the memory access time is a major part of the elapsed time compared with the floating point calculations part and therefore is …