A GPU-Based Kalman Filter for Track Fitting

Abstract

Computing centres, including those used to process High-Energy Physics data and simulations, are increasingly providing significant fractions of their computing resources through hardware architectures other than x86 CPUs, with GPUs being a common alternative. GPUs can provide excellent computational performance at a good price point for tasks that can be suitably parallelized. Charged particle (track) reconstruction is a computationally expensive component of HEP data reconstruction, and thus needs to use available resources in an efficient way. In this paper, an implementation of Kalman filter-based track fitting using CUDA and running on GPUs is presented. This utilizes the ACTS (A Common Tracking Software) toolkit; an open source and experiment-independent toolkit for track reconstruction. The implementation details and parallelization approach are described, along with the specific challenges for such an implementation. Detailed performance benchmarking results are discussed, which show encouraging performance gains over a CPU-based implementation for representative configurations. Finally, a perspective on the challenges and future directions for these studies is outlined. These include more complex and realistic scenarios which can be studied, and anticipated developments to software frameworks and standards which may open up possibilities for greater flexibility and improved performance.

Detray: a compile time polymorphic tracking geometry description

“…The project has investigated track based parallelization in a predefined telescope geometry and found a speedup of up to 4.6 in events with more than 1000 tracks compared to a multithreaded CPU implementation, while resorting to a custom matrix inversion algorithm. For details, please see [3].…”

Scientific Computing Plan for the ECCE Detector at the Electron Ion Collider

“…The role of hybrid architectures should also be considered in the ECCE reconstruction framework. Specifically, the use of GPU architectures will be important both for integrating machine learning into reconstruction workflows as well as generically taking advantage of the significant computational speed improvements that GPUs can provide, for example in charged particle reconstruction [15]. This integration has the added benefit of potentially utilizing the various leadership computing facilities that are available at national laboratories around the country, for more see Section 4.…”

Machine learning for track reconstruction at the LHC

“…Two complimentary strategies towards achieving efficient fullfledged KF tracking on the GPU are explored in ref. [6]. The first strategy revolves around running many track fits concurrently within an event, using CUDA threads to manage the parallelization.…”

“…The framework includes an ONNX [21] plugin, allowing the use of neural networks anywhere in the tracking workflow. R & D is also ongoing for support of accelerated hardware such as GPUs in ACTS [6]. The physics efficiency is defined as the fraction of all true particles that are reconstructed, while the technical efficiency is the fraction of all reconstructable particle (i.e.…”