Dr. Ezhilmathi Krishnasamy, University of Ljubljana, Faculty of Mechanical Engineering
Автор: Nacionalni kompetenčni center SLING
Загружено: 2025-12-04
Просмотров: 6
Directive-Based Programming Model for GPU Acceleration in Scientific Applications - Dr. Ezhilmathi Krishnasamy, University of Ljubljana, Faculty of Mechanical Engineering, LECAD Laboratory
While traditional GPU programming can be challenging, primarily because developers must learn low-level programming models designed for specific GPU architectures (such as CUDA for NVIDIA GPUs and HIP for AMD GPUs). In addition, there are alternative approaches with low-level or library-level programming models, to provide cross platform support on various GPUs. However, low-level programming still presents challenges, including unique syntax and methodologies associated with each programming model. Therefore, having an alternative programming model that emphasizes simplicity while maintaining good performance on both NVIDIA and AMD GPUs could be advantageous. This presentation will focus on how OpenACC and OpenMP Offloading can be effectively utilized for porting scientific applications to GPUs (both NVIDIA and AMD) without significant performance loss compared to traditional low-level programming models such as CUDA and HIP.
Доступные форматы для скачивания:
Скачать видео mp4
-
Информация по загрузке:
