Speaker
Description
Beam-beam simulation underpins luminosity prediction and detector-background estimation at FCC-ee. GuineaPig, the community-established tool used for these simulations, is single-threaded: a single FCC-ee Z bunch crossing takes several hours on one CPU core, making high-statistics Z-pole simulation campaigns very computationally expensive.
We present Guinea-Pig X (GPX), a CUDA port of GuineaPig that moves the bulk of the Particle-In-Cell (PIC) simulation steps onto the GPU. We will demonstrate how GPX achieves O(10x) improvement in simulation time compared to legacy GuineaPig, while providing statistically compatible outputs with GuineaPig. We will also present specific benchmarks of GPX on FCC-ee beam parameter scenarios, demonstrating the potential gains for the high-statistics campaigns required for luminosity-spectrum and beam-induced-background studies. Since each simulation occupies a single GPU, campaign throughput scales near-linearly with the number of available GPUs, making full-scale FCC-ee studies practical on a single multi-GPU node. We will close by discussing where the remaining bottlenecks lie and the outlook for further acceleration.
