Porting HEP Parameterized Calorimeter Simulation Code to GPUs.

Area of Science:

• High Energy Physics
• Computational Physics
• Accelerator Technology

Background:

• High Energy Physics (HEP) experiments, including those at the Large Hadron Collider (LHC), heavily rely on CPU resources for detector simulations and data analysis.
• The increasing data rates from LHC upgrades necessitate exploring alternative computing architectures beyond traditional CPUs.

Purpose of the Study:

• To investigate the feasibility of porting High Energy Physics simulation code to Graphics Processing Units (GPUs).
• To assess the potential for significant speed-ups in calorimeter simulations for future high-luminosity LHC conditions.

Main Methods:

• Porting the FastCaloSim, an ATLAS fast parametrized calorimeter simulation, to NVIDIA GPUs using CUDA.
• Developing a preliminary Kokkos implementation for broader architectural portability.

Main Results:

• Demonstrated the feasibility of GPU parallelization for calorimeter simulations.
• Achieved significant speed-ups in processing large datasets at both particle and event levels.
• Identified benefits for high-luminosity LHC scenarios with high per-event particle multiplicities.

Conclusions:

• GPU acceleration offers a viable solution to meet the growing computational demands of HEP experiments.
• Porting simulation tools like FastCaloSim to GPUs is essential for efficient data analysis at the high-luminosity LHC.
• Further development with Kokkos can enhance portability across diverse computing platforms.