No abstract

When dealing with nuclear reactor calculation schemes, the need for 3D transport-based reference solutions is essential for validation and optimization purposes. As SN transport methods may be considered promising with respect to comprehensive parallel computations, a 3D full PWR core benchmark was proposed to challenge the capabilities of the PENTRAN parallel SN code utilizing an IBM-BG/P computer. After a brief description of the benchmark, a parallel performance analysis is carried out, and shows that the parallelizable (Amdahl) fraction of PENTRAN is comprised between 0.994 ≤ f ≤ 0.996 for a number of BG/P nodes ranging from 17 to 1,156. The associated speedup reaches a value greater than 200 with 1,156 nodes. Using a best estimate model, PENTRAN results are then compared to Monte Carlo results rendered using the MCNP5 code. Good consistency is observed between the two methods (SN and Monte Carlo), with discrepancies less than 65 pcm for the k eff , and less than 2.5% for the flux at the pincell level.

No abstract

No abstract