SciPost Submission Page
Many-gluon tree amplitudes on modern GPUs: A case study for novel event generators
by Enrico Bothmann, Walter Giele, Stefan Hoeche, Joshua Isaacson, Max Knobbe
Submission summary
| As Contributors: | Enrico Bothmann · Max Knobbe |
| Arxiv Link: | https://arxiv.org/abs/2106.06507v2 (pdf) |
| Code repository: | https://www.gitlab.com/ebothmann/blockgen-archive |
| Date submitted: | 2022-01-11 14:45 |
| Submitted by: | Bothmann, Enrico |
| Submitted to: | SciPost Physics Codebases |
| Academic field: | Physics |
| Specialties: |
|
| Approach: | Computational |
Abstract
The compute efficiency of Monte-Carlo event generators for the Large Hadron Collider is expected to become a major bottleneck for simulations in the high-luminosity phase. Aiming at the development of a full-fledged generator for modern GPUs, we study the performance of various recursive strategies to compute multi-gluon tree-level amplitudes. We investigate the scaling of the algorithms on both CPU and GPU hardware. Finally, we provide practical recommendations as well as baseline implementations for the development of future simulation programs. The GPU implementations can be found at: https://www.gitlab.com/ebothmann/blockgen-archive.
Current status:
Author comments upon resubmission
List of changes
- Provide additional instructions for compiling and running the code in the linked code repository
- Add minor clarifications and improve some formulations
- Improve layout and figure placement, fix typographical/punctuation issues
- Add two missing references for multi-jet measurements at UA1 and UA2
- Clarify why it is advantageous to use a real-number only algorithm for gluon-only algorithms
- Add technical details on the GPU used for the study
- Improve results on the comparison between sampling and summing with respect to reaching a pre-defined precision goal; related to that, add a comment on the use of an improved helicity sampling algorithm
- Extend the appendix to make our point on a possible hybrid CPU/GPU ansatz clearer; in particular note that we can not compare to a GPU-only ansatz, as a full-fledged phase-space generator implementation is still missing