Speaker
Description
Significant computing resources are devoted to event generation for the Large Hadron Collider (LHC), with projected demands for the High-Luminosity LHC phases expected to increase substantially and go beyond the forecasted pledged resources. At the same time, High Performance Compute (HPC) clusters provide major compute resources which in many cases rely on the deployed GPU hardware.
Madgraph5_aMC@NLO and Sherpa/Pepper are parton-level event generator projects which recently have invested into hardware acceleration via offloading onto GPUs and CPU vector instructions. We present the current status of the projects, focussing on their CPU/GPU performance, scaling behaviour on modern HPC platforms, and their applicability to the production of computationally expensive background samples for LHC analyses. We further discuss recent developments enabling improved physics reach, including machine-learning-optimised phase-space integration using Normalizing Flow models, progress towards hardware-accelerated higher-order calculations, and physics-driven improvements to numerical stability in infrared-sensitive regions.
These advances illustrate how next-generation event generation can meet the combined challenges of precision, scalability, and sustainability for the HL-LHC and future collider experiments.
| Speaker confirmation | Yes |
|---|