On Improvement Of DEM-Implementation Based On CUDA

DEM(Discrete Element Method)is a numerical simulation method which is published by Cundall.It's widely used in the study and analysis of many particles in the process of interaction between the movements.The essential theory of DEM is the discrete motion of continuous object into non-continuous objects to work out the action status of an object.Fast computing and the need of less on-chip memory is the characteristic of DEM discrete element method.Our research group has a comprehensive understanding of DEM during many years' research and found that with the increasing of the number of particles during simulation,the results were not good since the simulation process costs too much time to work it out.To accelerate the calculation of whole simulation process,our research group proposed a solution,that was to use GPU instead of CPU,which has more core than CPU to compute the formula currently,the ratio can be 10 or 100,but less powerful performance in single core,which cannot handle a complicated logical calculation easily.At first,we utilized Open CL to speed up the simulation process.Open CL is a popular parallel computing software used by many researchers,which is not limited by special hardware.But it still has many shortcomings that it cannot provide doubleprecision calculation and high-level-currently-computing.CUDA is a parallel computing software published by NVIDIA Corporation.It provides more accurate computing such as double-precision computing than Open CL.Moreover,the SDK of CUDA provides lots of API for developers to make the programs easily.Based on the existing theory and algorithm,in this paper,we use the CUDA instead to meet the requirements of speed and accuracy.Moreover,based on the comparison between the results of CUDA and Open CL,we found that the time cost of CUDA is the same as Open CL,but CUDA provides much more accurate results.At the same time,in accord to the algorithm and the characteristics of the CUDA parallel computing architecture,the CUDA particle system is optimized to further speed up the operation.At last,we use lots of data to test the collision between particles of CUDA particle system to prove that it can provide the right results as the previous one.