Parton Showers on GPUs with GAPS

Upcoming LHC runs will not only push experimental precision — they will also push computing demands to new limits. Recent CERN studies indicate that simulation workloads are set to exceed the computing budget for Run 4. Monte Carlo event generators are one of the largest contributors to this increase, both for experimental and theoretical needs.

To address this challenge, we are exploring GPU acceleration for fast and energy-efficient event simulation. Our goal is to provide a practical alternative to CPU clusters while maintaining physics fidelity.

This effort has led to GAPS: a GPU-accelerated hard process and parton shower event generator. GAPS supports both CPU and GPU execution and has been benchmarked on LEP- and LHC-like simulations. We have demonstrated that a single modern GPU can simulate 1,000,000 events with comparable throughput and lower energy consumption than multi-node CPU clusters.

More details can be found in our publications:

📄 Publications

• M. H. Seymour & S. Sule, An Algorithm to Parallelise Parton Showers on a GPU,
  SciPost Physics Codebases 33 (2024)
  https://scipost.org/SciPostPhysCodeb.33

• M. H. Seymour & S. Sule, An NLO-Matched Initial and Final State Parton Shower on a GPU,
  arXiv:2511.19633 (2025)
  https://arxiv.org/abs/2511.19633

💻 Code

• GAPS — GPU-Amplified Parton Shower
  https://gitlab.com/siddharthsule/gaps