Speaker
Description
The emerging computable devices, graphical processing units (GPU), are gradually applied in the simulations of space physics for their high efficiency. Here we present a fully kinetic particle-in-cell (PIC) simulation code running on GPU devices called GPIC, which adopts CUDA Fortran programming. Compared with the simulations running on the Intel Xeon Gold processor, our program working on NVIDIA A100 can achieve amazing accelerated computing up to 724 times faster than the former. Besides, we also implement it on multiple GPUs across computing nodes and obtain relatively low latency. By applying this program in the simulations of magnetic reconnection, we find there exists a crater structure behind the reconnection front, which is well supported by the MMS observations. This structure is formed by the continuous impact of high-speed electron outflow jets. Our advanced program shows its capability and superiority in performing PIC simulations with low cost. The scientific discovery in the magnetic reconnection indicates a new energy conversion pattern between electrons and the magnetic field.
