GTC Silicon Valley-2019: GPUs for Data Acquisition and Simulation for the Muon g-2 Experiment at Fermilab
Note: This video may require joining the NVIDIA Developer Program or login
GTC Silicon Valley-2019 ID:S9237:GPUs for Data Acquisition and Simulation for the Muon g-2 Experiment at Fermilab
Wesley Gohn(Siemens Healthineers),Ran Hong(Argonne National Lab)
Learn how we're using GPUs to accelerate the data-acquisition rate and provide fast simulations for a new particle physics experiment, Muon g-2 at Fermilab. The experiment involves measuring the anomalous magnetic moment of the muon. We'll discuss how we're acquiring data at a raw data rate 20GB per second and processing it in real time with CUDA and 28 NVIDIA Tesla K40 GPUs. This reduces the rate of data written to disk by a factor of 100. We'll also cover how we're using NVIDIA GPUs to accelerate online FFTs and linear fits in magnetic-field measurement. Because the experiment will measure 150 billion muon decays, we must perform very fast simulations to fully understand the data. Conventional computing is impractical for these simulations. Instead, we're using a CUDA-based Monte Carlo code running on GPU nodes on the Open Science Grid. We'll describe how the simulation can generate the entire three-year dataset in less than three hours.