Author(s)

Haiyi Ye, Ching-Hsing Luo, Xinji Chen

Corresponding Author

Haiyi Ye

ABSTRACT

Cardiac electrophysiological simulation is a very complex computational process. Running on Graphics processing unit (GPU) is an effective method for cardiac electrophysiological simulation. In addition, the use of adaptive time-step can also effectively speed up the simulation of heart cells. However, the previous works running on GPU could not get apparent speedup (only 1.5 times). This paper implements adaptive time-step methods on multi-GPU to run Luo-Rudy dynamic (LRd) ventricular action potential model incorporating a Markov Sodium Channel model. For adaptive time-step methods, we use Traditional Hybrid Method(THM) and Chen-Chen-Luo’s quadratic adaptive algorithm (CCL). As LRd is solved by THM or CCL in a single GPU, the acceleration is 17.5 times(17.5x) and 43x respectively compared with the fixed time-step under Mix Root Mean Square error lower than 5%. In 2 GPUs, the acceleration of THM and CCL is 9.7x and 33.7x separately. As there are 4 GPUs, the acceleration of THM is only 5.7 x while acceleration is 25.6x for CCL. In addition, compared with the fixed time-step in CPU, THM and CCL accelerated to 1861x and 5054x in a single GPU, 1885x and 6652x in 2 GPUs, 2093x and 8024x in 4 GPUs. In conclusion, CCL is much better than THM to save computation cost on multi-GPU, while the speedup is not lineally increased with the number of GPU.

KEYWORDS

Computer Simulation, Ventricular Cell, Adaptive Time-step Method, Graphics Processing Unit, High Performance Computing