Triggering events with GPUs at ATLAS

21st International Conference on Computing in High Energy and Nuclear Physics

物理学;计算机科学

Kama, S.^1 ; Soares, J Augusto^2,6 ; Baines, J.^3 ; Bauce, M.^4 ; Bold, T.^5 ; Muino, P Conde^2 ; Emeliyanov, D.^3 ; Goncalo, R.^6 ; Messina, A.^4 ; Negrini, M.^7 ; Rinaldi, L.^8 ; Sidoti, A.^7 ; Delgado, A Tavares^6 ; Tupputi, S.^9 ; Lopes, L Vaz Gil^6

Southern Methodist University, Dallas

TX, United States^1

INESC-ID and Faculdade de Ciencias, Universidade de Lisboa, Portugal^2

STFC and Rutherford Appleton Lab, United Kingdom^3

Sapienza Universita di Roma and INFN, Italy^4

AGH Univ. of Science and Technology, Krakow, Poland^5

LIP, Lisboa, Portugal^6

Istituto Nazionale di Fisica Nucleare (INFN), Italy^7

Universita di Bologna and INFN, Italy^8

INFN-CNAF, Italy^9

关键词: Complex particles;    Computing power;    High-level triggers;    Integration procedure;    Massively parallels;    Online selection;    Reconstruction algorithms;    Reduced power consumption;   

Others  :  https://iopscience.iop.org/article/10.1088/1742-6596/664/9/092014/pdf
DOI  :  10.1088/1742-6596/664/9/092014

学科分类：计算机科学（综合）

来源: IOP

【 摘 要 】

The growing complexity of events produced in LHC collisions demands increasing computing power both for the online selection and for the offline reconstruction of events. In recent years there have been significant advances in the performance of Graphics Processing Units (GPUs) both in terms of increased compute power and reduced power consumption that make GPUs extremely attractive for use in a complex particle physics experiments such as ATLAS. A small scale prototype of the full ATLAS High Level Trigger has been implemented that exploits reconstruction algorithms optimized for this new massively parallel paradigm. We discuss the integration procedure followed for this prototype and present the performance achieved and the prospects for the future.