Electron tracks simulation in water: Performance comparison between GPU CPU and the EUMED grid installation.

Area of Science:

• Medical Physics
• Computational Physics
• Radiation Physics

Background:

• Monte Carlo methods are crucial for simulating particle transport in biological tissues.
• Geant4-DNA provides a toolkit for simulating interactions of low-energy electrons and protons.
• Minimizing simulation time is essential for complex biological modeling.

Purpose of the Study:

• To evaluate and compare the efficiency of different computational approaches for electron track generation simulations.
• To identify the optimal technology (CPU, GPU, or computing grid) for minimizing simulation time in Geant4-DNA electron transport.

Main Methods:

• Development of a Graphics Processing Unit (GPU) software tool for electron track simulations.
• Implementation of a Central Processing Unit (CPU) version using identical collision models.
• Performance comparison of CPU, GPU, and computing grid (France Grilles) simulations against Geant4-DNA processes.

Main Results:

• CPU simulations are efficient for fewer than 10^4 electrons at 100 eV, with optimal numbers decreasing at higher energies.
• GPU simulations excel with over 10^4 electrons at 100 eV, down to 10^3 at 10 keV, showing increased efficiency with higher energy.
• Computing grid simulations are most effective for over 10^5 electrons at 10 keV, with performance benefits increasing with electron count and energy, despite queuing delays.

Conclusions:

• Central Processing Unit (CPU) is suitable for simulations involving a small number of primary electrons.
• Graphics Processing Unit (GPU) is advantageous when sufficient particles occupy GPU resources, maximizing computational power.
• Computing grids are best for large-scale simulations requiring high computational power for numerous primary electrons at high energies.