OuroborosBEM: a fast multi-GPU microscopic Monte Carlo simulation for gaseous detectors and charged particle dynamics

The design of gaseous detectors for accelerator, particle and nuclear physics requires simulations relying on multi-physics aspects. In fact, these simulations deal with the dynamics of a large number of charged particles interacting in a gaseous medium immersed in the electric field generated by a more or less complex assembly of electrodes and dielectric materials. We report here on a homemade software, called ouroborosbem, able to tackle the different features involved in such simulations. After solving the electrostatic problem for which a solver based on the boundary element method (BEM) has been implemented, particles are tracked and will microscopically interact with the gas medium. Dynamical effects have been included such as the electron-ion recombination process, the charging-up of the dielectric materials and other space charge effects that might alter the detector performances. These were made possible thanks to the nVidia CUDA language specifically optimised to run on Graphical Processor Units (GPUs) to minimize the computing times. Comparisons of the results obtained for parallel plate avalanche counters and GEM detectors to literature data on swarm parameters fully validate the performances of ouroborosbem. Moreover, we were able to precisely reproduce the measured gains of single and double GEM detectors as a function of the applied voltage.

Acceleration of boundary element calculations for closed domain using nonlinear form functions and CUDA technology

The evolution of computer technologies, as a hardware and a software parts, allows to attain fast and accurate solutions to many applied problems in scientific areas. Acceleration of calculations is broadly used technic that is basically implemented by multithreading and multicore processors. NVidia CUDA technology or simply CUDA opens a way to efficient acceleration of boundary elements method (BEM), that includes many independent stages. The main goal of the paper is implementation and acceleration of indirect boundary element method using three form functions. Calculation of the potentialdistribution inside a closed boundary under the action of the defined boundary condition is considered. In order to accelerate corresponding calculations, they were parallelized at the graphic accelerator using NVidia CUDA technology. The dependences of acceleration of parallel computations as compared with sequential ones were explored for different numbers of boundary elements and computational nodes. A significant acceleration (up to 52 times) calculation of the potential distribution without loss in accuracy is shown. Acceleration of up to 22 times was achieved in calculation of mutual influence matrix for boundary elements. Using CUDA technology allows to attain significant acceleration without loss in accuracy and convergence. So application of CUDA is a good way to parallelizing BEM. Application of developed approach allows to solve problems in different areas of physics such as acoustics, hydromechanics, electrodynamics, mechanics of solids and many other areas, efficiently.

GPU-accelerated molecular dynamics: State-of-art software performance and porting from Nvidia CUDA to AMD HIP

Classical molecular dynamics (MD) calculations represent a significant part of the utilization time of high-performance computing systems. As usual, the efficiency of such calculations is based on an interplay of software and hardware that are nowadays moving to hybrid GPU-based technologies. Several well-developed open-source MD codes focused on GPUs differ both in their data management capabilities and in performance. In this work, we analyze the performance of LAMMPS, GROMACS and OpenMM MD packages with different GPU backends on Nvidia Volta and AMD Vega20 GPUs. We consider the efficiency of solving two identical MD models (generic for material science and biomolecular studies) using different software and hardware combinations. We describe our experience in porting the CUDA backend of LAMMPS to ROCm HIP that shows considerable benefits for AMD GPUs comparatively to the OpenCL backend.

RECENT CAPABILITY AND PERFORMANCE ENHANCEMENTS OF THE WHOLE-CORE TRANSPORT CODE nTRACER

The whole-core transport code nTRACER has made many advances in recent years. Several innovative cross section treatment methods were developed, a new axial transport solver was introduced for stabilizing the 2D/1D scheme, and substantial computational enhancements were achieved using NVIDIA CUDA and Intel Math Kernel Library (MKL). In addition, gamma transport solver was implemented to predict the power distributions more physically, and the flexibility of the restart calculation was improved using an offline processing code nTIG (nTRACER Input Generator). This paper is the compilation of the recent progresses in nTRACER developments.

COMPARISON OF GPU AND CPU EFFICIENCY WHILE SOLVING HEAT CONDUCTION PROBLEMS

Overview of GPU usage while solving different engineering problems, comparison between CPU and GPU computations and overview of the heat conduction problem are provided in this paper. The Jacobi iterative algorithm was implemented by using Python, TensorFlow GPU library and NVIDIA CUDA technology. Numerical experiments were conducted with 6 CPUs and 4 GPUs. The fastest used GPU completed the calculations 19 times faster than the slowest CPU. On average, GPU was from 9 to 11 times faster than CPU. Significant relative speed-up in GPU calculations starts when the matrix contains at least 4002 floating-point numbers.

Aplicação de Evolução Diferencial em GPU Para o Problema de Predição de Estrutura de Proteínas com Modelo 3D AB Off-Lattice

A função que uma proteína exerce está diretamente relacionada com a sua estrutura tridimensional. Porém, para a maior parte das proteínas atualmente sequenciadas ainda não se conhece sua forma estrutural nativa. Este artigo propõe a utilização do algoritmo de Evolução Diferencial (DE) desenvolvido na plataforma NVIDIA CUDA aplicado ao modelo 3D AB Off-Lattice para Predição de Estrutura de Proteínas. Uma estratégia de nichos e crowding foi implementada no algoritmo DE combinada com técnicas de autoajuste de parâmetros, rotinas para reinicialização da população, dois níveis de otimização e busca local. Quatro proteínas reais foram utilizadas para experimentação e os resultados obtidos se mostram competitivos com o estado-da-arte. A utilização de paralelismo massivo através da GPU ressalta a aplicabilidade desses recursos a esta classe de problemas atingindo acelerações de 708.78x para a maior cadeia proteica.