Paper VT-TuM12
Introduction to Molflow+, a New GPU-Based Monte Carlo Code for Simulating Molecular Flows and for Calculating Angular Coefficients in the CUDA Environment

Tuesday, October 21, 2008, 11:40 am, Room 205

Molflow+ is a new test-particle (TP) monte carlo (MC) code developed for the simulation of molecular flows. It is the natural evolution of the code Molflow which had been developed by one of the authors (R.K.) and used by many laboratories since 1991. Molflow+ implements modern computing trends, such as the use of the OpenGL graphical interface and C\C++ code, and takes advantage of recent developments in the field of graphical processing units (GPUs) which allow a substantial parallelization of the TP MC algorithm for a modest capital cost. The Compute Unified Device Architecture (CUDA) environment has been chosen for its ease of use and portability to different computing environments. A second version of the program will let a similar code run on computers which have no GPUs or non CUDA-compatible GPUs, and only single or multi-core CPUs. As for the original code Molflow, in Molflow+ the geometry of the vacuum component under study is described in terms of planar polygonal surfaces to which attributes such as transparency, sticking coefficient, desorption profiles and more are assigned. The geometry is defined either by using a built-in editor program or by importing files obtained by using a popular computer-assisted design program which generates a triangularization of the surfaces. A very large period random number generator has been implemented into the code, in order to minimize systematic statistical errors. It will be shown how Molflow+ can also be used to calculate the angular coefficients of arbitrarily complex geometries, which can then be used to perform vacuum calculations based on the resolution of matrix operations. It will also be shown how using the GPU yields a dramatic improvement of the processing speed, thus allowing the simulation of complex vacuum systems which could not hitherto be carried out unless powerful and expensive computing tools were used. The paper will outline the early stages of development of the code, its benchmarking against existing analytical as well as numerical results, and will hint at possible fields of application where the dramatic computational power of the GPU could make a difference as compared to CPU-only cases. It will also discuss possible future developments such as the inclusion of intramolecular collisions which could extend the domain of application to flow regimes other than molecular.