Present Status and Extensions of the Monte Carlo Performance Benchmark
1 Delft Nuclear Consultancy, IJsselzoom 2, 2902 LB Capelle aan den IJssel, The Netherlands
2 Georgia Institute of Technology, Nuclear and Radiological Engineering, Atlanta, GA 30332-0405, USA
3 University of Michigan, Department of Nuclear Engineering and Radiological Sciences, Ann Arbor, MI 48109-2104, USA
* Corresponding Author, E-mail: email@example.com
The NEA Monte Carlo Performance benchmark started in 2011 aiming to monitor over the years the abilities to perform a full-size Monte Carlo reactor core calculation with a detailed power production for each fuel pin with axial distribution. This paper gives an overview of the contributed results thus far. It shows that reaching a statistical accuracy of 1 % for most of the small fuel zones requires about 100 billion neutron histories. The efficiency of parallel execution of Monte Carlo codes on a large number of processor cores shows clear limitations for computer clusters with common type computer nodes. However, using true supercomputers the speedup of parallel calculations is increasing up to large numbers of processor cores. More experience is needed from calculations on true supercomputers using large numbers of processors in order to predict if the requested calculations can be done in a short time.
As the specifications of the reactor geometry for this benchmark test are well suited for further investigations of full-core Monte Carlo calculations and a need is felt for testing other issues than its computational performance, proposals are presented for extending the benchmark to a suite of benchmark problems for evaluating fission source convergence for a system with a high dominance ratio, for coupling with thermal-hydraulics calculations to evaluate the use of different temperatures and coolant densities and to study the correctness and effectiveness of burnup calculations. Moreover, other contemporary proposals for a full-core calculation with realistic geometry and material composition will be discussed.
Key words: Monte Carlo / benchmark / performance / parallel / power distribution
© Owned by the authors, published by EDP Sciences, 2014