Algorithm Acceleration from GPGPUs for the ATLAS Upgrade

The future upgrades to the LHC are expected to increase the design luminosity by an order of magnitude leading to new computational challenges for the ATLAS experiment. One such challenge will be the ability to handle a much higher rate of interesting physics events by the ATLAS High Level Trigger system. We will present results from the adoption of General Purpose Graphics Processing Units (GPGPUs) to provide computational acceleration for key algorithms in the ATLAS Inner Detector Trigger. The z-finder algorithm - used to determine the accurate z position of primary interactions - and the Kalman Filter based track reconstruction routine have been adapted for GPGPU execution using the CUDA parallel computing architecture. We describe the programming and benchmarking methods used and demonstrate the relative throughput performance for different trigger scenarios. Where significant performance boost is found we will outline how GPGPU acceleration could be exploited and incorporated into the future ATLAS computing framework