Fig. 13

Global performance varying with grid sizes. The speedup is defined as the ratio of elapsed time of running on CPU with one core to lapsed time of running with a GPU. The accelerating performance of three GPUs is in agreement with their DP performance (see Table 1). Titan V gets the highest speedup, due to maximum number of CUDA cores and maximum number of double precision processing units. Although RTX 2080 Ti has more CUDA cores than Tesla P100, it has less double precision processing units than Tesla P100. When the grid size increases, speedups of the three GPUs increase in general. But Titan V and Tesla P100 get higher increments. When the grid size is larger than 15 million, Tesla P100 and RTX 2080 Ti almost reach their limits, while there is space for Titan V