DPNC-HEPIA Heterogeneous Computing System
A high-performing heterogeneous computing system has been set up in collaboration between DPNC and HEPIA. The system was originally equipped with a GPU-accelerated server hosting two AMD VCK5000 FPGA devices. A second server, equipped with AMD U55C and V80 FPGAs, was subsequently added. Both systems support R&D for heterogeneous computing in ATLAS.1
The infrastructure was further complemented by a powerful NVIDIA Grace Hopper GPU system running NVIDIA Holoscan, used for R&D on the data-acquisition system for CTAO cameras in a collaboration between DPNC and HEPIA.2
The resulting heterogeneous computing infrastructure, bringing together the ATLAS and CTAO R&D platforms through a common high-speed network, is illustrated below.
Projects using the Heterogeneous Computing System
In the context of ATLAS Event-Filter / algorithmic acceleration
within the research groups of A. Sfyrla (UniGe) and/or Q. Berthet and A. Upegui (HEPIA)- AthXRT: Centralized FPGA Management for Accelerated Algorithms in Athena (Q. Berthet, ATL-DAQ-PROC-2025-002, presented at CHEP 2024)
- QDIPS: Deep Sets Network for FPGA investigated for high speed inference on ATLAS (C. Antel, ATL-DAQ-PROC-2025-008, presented at CHEP 2024; ongoing)
- Topocluster acceleration (A. Marquet, to be published)
- CaloJetSSD implementation to FPGA (A. Marquet and D. Fiacco, in progress)
- Doctoral thesis (Kostas Axiotis, CERN and UniGe PhD student, 2020–2025)
- Topo-Automaton Clustering acceleration on FPGA (Kirill Goundiaev, Projet de Master HEPIA 2025)
- GPU Implementation of the Ringer Algorithm in the ATLAS High-Level Trigger (Quentin Saucy, Projet de Master HEPIA 2026)
- Intégration du décodage de donnée d’ATLAS sur V80 avec Coyote (Abivarman Kandiah, Projet d’Approfondissement HEPIA 2026)
- Étude d'implémentation de réseaux de neurones U-NET sur FPGA à l'aide de HLS4ML (Vincent Andrey, Projet de Bachelor HEPIA 2023)
- Déploiement sur FPGA d’un modèle de machine learning pour le système de trigger d’ATLAS (Gaspard Le Gouic, Projet de Bachelor HEPIA 2025)
- Implémentation FPGA d’un modèle convolutif pour la reconstruction de jets dans ATLAS (Michael Divia, Projet de Bachelor HEPIA 2025)
- Intégration d’un framework de calcul hétérogène dans l’expérience ATLAS (Arlind Baftiju, Projet de Bachelor HEPIA 2026)
- Accélération de calcul pour le décodage de données du sous-détecteur "Tile calorimeter" (Tom Andrivet, Projet de Bachelor HEPIA 2026)
In the context of CTAO
- The Cherenkov Telescope Array Observatory: Status, Science and real-time processing, M. Heller at ODISSEE All-Hands meeting.
1 The ATLAS systems were co-funded by the
SNF FLARE grant and HEPIA.
2 The CTAO system was supported by SERI measure A231.0287.