An overview of Celeritas: a novel GPU Monte Carlo detector simulation code
ORAL
Abstract
The next-generation of High Energy Physics (HEP) experiments will rely on a vast increase in detector complexity and data collection, leading to an unprecedented amount of computing storage and processing capacity needs. Contemporary increases in computing capacity are primarily due to the use of heterogeneous architectures that rely on the high performance-per-Watt of GPUs. Celeritas, a new GPU-optimized detector simulation code, seeks to unlock the resources of DOE's Leadership Computing Facilities (LCFs) and the next generation of computing grid hardware for HEP experiments. Early results show a 40× speedup factor for standalone EM simulations on LCF computers when using GPUs. This talk will provide an overview of Celeritas, focusing on its physics capabilities and integration with the CPU-based detector simulation code Geant4. Current Celeritas' electromagnetic (EM) physics models for electrons, positrons, and photons will be verified against Geant4, with an additional verification for coupled offloading, where Geant4 processes hadronic and decay physics but sends EM particles to Celeritas, with an estimated speedup of 2–3× compared to a Geant4-only run. The presentation will also preview future plans for integration with HEP experiments, of which EM offloading is the first step.
–
Presenters
-
Stefano C. Tognini
Oak Ridge National Laboratory
Authors
-
Stefano C. Tognini
Oak Ridge National Laboratory
-
Seth R Johnson
Oak Ridge National Laboratory
-
Soon Yung Jun
Fermi National Accelerator Laboratory
-
Amanda L Lund
Argonne National Laboratory
-
Thomas M Evans
Oak Ridge National Laboratory
-
Philippe Canal
Fermi National Accelerator Laboratory
-
Paul K Romano
Argonne National Laboratory
-
Guilherme Lima
Fermi National Accelerator Laboratory