Exascale Plasma-Accelerator Simulations with PIConGPU

Debus, Alexander; Kluge, Thomas; Widera, René; Bastrakov, Sergei; Steiniger, Klaus; Garten, Marco; Lebedev, Anton; Pausch, Richard; Meyer, Felix; Pöschel, Franz; Kelling, Jeffrey; Juckeland, Guido; Stephan, Jan; Herten, Andreas; Chandrasekaran, Sunita; Leinhauser, Matthew; Young, Jeff; Davis, Joshua H.; Diaz, Jose M.; Huebl, Axel; Hernandez, Benjamin; Chatterjee, Ronnie; Rogers, David; Bussmann, Michael

Exascale Plasma-Accelerator Simulations with PIConGPU

Debus, A.; Kluge, T.; Widera, R.; Bastrakov, S.; Steiniger, K.; Garten, M.; Lebedev, A.; Pausch, R.; Meyer, F.; Pöschel, F.; Kelling, J.; Juckeland, G.; Stephan, J.; Herten, A.; Chandrasekaran, S.; Leinhauser, M.; Young, J.; Davis, J. H.; Diaz, J. M.; Huebl, A.; Hernandez, B.; Chatterjee, R.; Rogers, D.; Bussmann, M.

* Next-generation Laser-plasma Accelerators (protons and electrons) require fast and predictive 3D particle-in-cell simulations at the exascale.
* PIConGPU is performance portable & available as a single source through Alpaka.
* PIConGPU scales to the largest supercomputers of the world, leveraging Alpaka for running on the latest GPUs from NVIDIA (A100) and AMD (MI100)
* HIP & OpenMP 5+ backends now available on Alpaka and PIConGPU.
* PIConGPU performance scales on Juwels Booster and Summit machines.

Starting from typical simulation applications and requirements in laser-plasma accelerators, we present the PIConGPU software stack and current scaling results on large-scale compute clusters. We show results from the Frontier CAAR project and the JUWELS Booster early-access project.

Keywords: PIC simulation; PIConGPU; exascale computing; HPC; laser-plasma physics; Alpaka; performance portability

Lecture (Conference) (Online presentation)
7th MT Meeting, 01.-3.2.2021, Virtuell, Deutschland

Permalink: https://www.hzdr.de/publications/Publ-33821