Peculiarities of exaflops computing in plasma physics

Abstract

The use of exaflops computing in plasma physics is necessary when the high-accuracy results are required in the case of nonequilibrium plasma. An approach is proposed to reach the exaflops performance by means of hybrid supercomputers with the particle-in-cell method in the numerical plasma physics. At present, performance is 0.8 Teraflops per one Tesla 2090 at the particle pushing stage. It means that the performance of the Tianhe-1A supercomputer with the PIC method may be 5.6 Petaflops (twice more than with LinPack). Therefore, the exaflops performance may be achieved with hybrid supercomputers faster than with classical supercomputers. It means that the 3D nonequilibrium PIC plasma computations with high resolution may be performed. Nevertheless, it is possible that the present data storage and transmission systems may not allow one to store, process and visualize the corresponding data. An estimate is given for the data size in plasma physics problems and also for the performance of data processing systems.