Utilisation of GPUs for the ATLAS trigger software and implementation of machine-learning algorithms for muon reconstruction in the ATLAS High-Level Trigger

The large increase in luminosity planned for the High Luminosity LHC gives rise to many challenges for the trigger and data acquisition system.  An instantaneous luminosity up to 7 × $10^{34} cm^{−2}s^{−1}$ is expected, corresponding to an average
number of inelastic $\rho \rho $ collisions per bunch crossing of around 200. For such data taking conditions, due to the unprecedented number of particle hits in the tracker system, ATLAS is putting effort to include multithread computation devices in the trigger architecture. Graphics Processing Units are one of the technologies under investigation. Moreover, track reconstruction algorithms must have stable performance with respect to pileup, in order to ensure the computing requirements of the experiment. In this note a study of the performance of a fast track reconstruction algorithm with the new ATLAS Inner Traker geometry is presented.