KCETA funded heterogeneous computing platform for reconstruction and trigger development

Funding from KCETA was used to set up the heterogeneous computing platform DeepThought2 at the Institute of Experimental Particle Physics (ETP). The system is jointly operated with the Institut für Technik der Informationsverarbeitung (ITIV) and serves as a shared infrastructure for interdisciplinary research between electrical engineering and experimental particle physics.

DeepThought2 is designed as a development and benchmarking platform for both training of machine learning projects and development of latency critical algorithms. It supports quantitative performance studies for ultrafast detector reconstruction across CPU, GPU, and FPGA architectures. It provides one terabyte of main memory and combines NVIDIA GPUs with an AMD Alveo V80 FPGA accelerator. It is used by students and researchers to develop and evaluate machine learning, reconstruction, and trigger algorithms for CMS and Belle II across heterogeneous computing technologies.