Development of GPU-Accelerated Particle Tracking Code MBTRACK2-CUDA for Studies of Collective Effects in Circular Accelerators

Keon-hee Kim (GANIL)

Particle tracking simulations are widely used to study collective beam instabilities in circular accelerators. Conventional approaches based on Central Processing Units (CPUs) perform single- or multi-bunch tracking using single-core or parallel multi-core computation with MPI and OpenMP. While these methods scale well for single-bunch studies, extending to multi-bunch tracking drastically increases computational demand, often requiring large CPU clusters. To address this, General-Purpose computing on Graphics Processing Units (GPGPU) enables tracking on a stand-alone PC. However, frequent CPU-GPU interactions, including data transfers and synchronization operations during tracking, can introduce communication overhead, potentially reducing the overall effectiveness of GPU-based computations. In this talk, a fully GPU-resident approach is presented, eliminating such communication by performing the entire tracking simulation on the GPU. The developed code, MBTRACK2-CUDA, is a CUDA-based port of MBTRACK2 designed for efficient single- and multi-bunch simulations. Its performance and validation are discussed, along with applications to resistive-wall instability studies in the HEPS and CEPC storage rings.