UPCommons. Portal del coneixement obert de la UPC
Publicación
Contribución de congreso (ConferenceOutput)
A qualifiable GPU sharing approach for AI workloads in critical systems
© 2025 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes,creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works., Graphics Processing Units (GPUs) are increasingly considered for adoption in safety-critical systems and especially space, in order to enable high performance on-board processing, particularly needed for Artificial Intelligence (AI) acceleration. While use of GPUs has been demonstrated in qualifiable real-time operating systems, GPU sharing has not been explored so far. In this work we describe we describe the design, implementation and evaluation of such GPU sharing approach, which is portable among different operating systems. The proposed solution has been implemented and demonstrated with two real-world AI applications for space missions. The results demonstrate significant performance improvements and efficient AI utilization. By addressing the timing constraints and enabling GPU utilization, this approach opens up new possibilities for efficient and reliable AI-powered space systems., The research presented in this paper has received funding by the European Community’s Horizon Europe programme under the METASAT project (grant agreement 101082622) and the Chips Act Joint Undertaking project SMARTY (grant agreement 101140087). Moreover, it has been supported by the project PID2024-155745OB-I00 funded by Spanish MICIU/AEI/10.13039/501100011033 / FEDER, UE (IAFE)., Peer Reviewed