GPU and FPGA-accelerated AI inference as a Service for Particle Physics

Shih-Chieh Hsu^1,2*

¹Physics, University of Washington, Seattle, USA
²Physics, National Tsing Hua University, Hsinchu, Taiwan

* Presenter:Shih-Chieh Hsu, email:schsu@uw.edu

New heterogeneous computing paradigms on dedicated hardware with increased parallelization, such as Graphics Processing Units (GPUs) and Field Programmable Gate Arrays (FPGAs), offer exciting solutions with large potential gains. The growing applications of Artificial Intelligence (AI) algorithms in particle physics for simulation, reconstruction, and analysis are naturally deployed on such platforms. We demonstrate that the acceleration of AI inference as a service represents a heterogeneous computing solution for particle physics experiments that potentially requires minimal modification to the current computing model. A series of workflows are developed to establish the performance capabilities of GPUs and FPGAs as a service. Multiple different devices and a range of algorithms for use in high energy physics are studied. For a small, dense network, the throughput can be improved by an order of magnitude with respect to GPUs as a service. For large convolutional networks, the throughput is found to be comparable to GPUs as a service. We discuss a strategy for the seamless integration of coprocessors so that the LHC can maintain, if not exceed, its current performance throughout its running. This work also represents the first open-source FPGAs-as-a-service toolkit. (Reference: https://arxiv.org/abs/2007.10359, https://arxiv.org/abs/2010.08556)

Keywords: Artificial Intelligence, Computing, GPU, FPGA, Accelerator