GTC 2020: High-Throughput Real-time Data Processing with GPUs at CERN
After clicking “Watch Now” you will be prompted to login or join.
Click “Watch Now” to login or join the NVIDIA Developer Program.
High-Throughput Real-time Data Processing with GPUs at CERN
Daniel Hugo Cámpora Pérez, NIKHEF and University of Maastricht
We'll present the design and performance considerations and system optimization of a GPU-based, real-time physics selection system at a Large Hadron Collider experiment. Millions of particles collide every second at the LHCb experiment at CERN in Switzerland. To select interesting particle collisions, data must pass through an acquisition system and be filtered with real-time selection software. The throughput processed in the first stage of this streaming data processing application amounts to 40 terabytes per second, and the efficiency of the selection is crucial toward improving our fundamental understanding of the universe. In order to process this massive data throughput in real-time, we developed GPU physics reconstruction software called Allen. The Allen framework hands the raw data to GPU streams, which perform the decoding, reconstruction, and selection of particle collisions.