Multi-threading performance of Geant4, MCNP6, and PHITS Monte Carlo codes for tetrahedral-mesh geometry

Han, M. C.*; Yeom, Y. S.*; Lee, H. S.*; Shin, B.*; Kim, C. H.*; Furuta, Takuya   

The multi-threading computation performances of the Geant4, MCNP6, and PHITS codes were evaluated using three tetrahedral-mesh phantoms with different complexity. Photon and neutron transport simulations were conducted and the initialization time, calculation time, and memory usage were measured as a function of the number of threads N used in the simulation. The initialization time significantly increases with the complexity of the phantom, but not much with the number of the threads. For the calculation time, Geant4 showed good parallelization efficiency with multi-thread computation (30 times speed-up factor for N = 40) adopting the private tallies while saturation of the speed-up factor were observed in MCNP6 and PHITS (10 and a few times for N = 40) due to the time delay for the sharing tallies. On the other hand, Geant4 requires larger memory specification and the memory usage rapidly increases with the number of threads compared to MCNP6 or PHITS. It is notable that when compared to the other codes, the memory usage of PHITS is much smaller, regardless of both the complexity of the phantom and the number of the threads.