Μελέτη επίδοσης εφαρμογών υψηλής έντασης σε ετερογενείς αρχιτεκτονικές

Μάρκου, Αθανάσιος; Markou, Athanasios

dc.contributor.author	Μάρκου, Αθανάσιος	el
dc.contributor.author	Markou, Athanasios	en
dc.date.accessioned	2022-01-10T11:59:22Z
dc.date.available	2022-01-10T11:59:22Z
dc.identifier.uri	https://dspace.lib.ntua.gr/xmlui/handle/123456789/54293
dc.identifier.uri	http://dx.doi.org/10.26240/heal.ntua.21991
dc.rights	Αναφορά Δημιουργού-Μη Εμπορική Χρήση-Όχι Παράγωγα Έργα 3.0 Ελλάδα	*
dc.rights.uri	http://creativecommons.org/licenses/by-nc-nd/3.0/gr/	*
dc.subject	Heterogeneous architectures	el
dc.subject	High performance computing	en
dc.subject	Cloud computing	en
dc.subject	Performance modeling	en
dc.subject	Power efficiency	en
dc.title	Μελέτη επίδοσης εφαρμογών υψηλής έντασης σε ετερογενείς αρχιτεκτονικές	el
heal.type	bachelorThesis
heal.classification	Parallel processing systems	en
heal.language	el
heal.access	free
heal.recordProvider	ntua	el
heal.publicationDate	2021-06-23
heal.abstract	While developing their parallel high-intensity applications researchers come across a crucial dilemma. The dilemma is whether their application should be targeted for a CPU- architecture or a GPU-architecture. A common strategy for researchers is to first implement an OpenMP version of their application to have a quantitative estimation of how well their algorithm leverages from the use of a multicore CPU-node. If the performance attained is not sufficient one has to reconsider his/her approach. Tuning perfectly your OpenMP code to utilize special architectural features of the target CPU-node might not always be the solution that yields the optimal performance, maybe another architecture is more suitable for leveraging from the application’s scalability and thus yielding the optimal performance. Maybe one GPU is more suitable for this applicaton. An additional concern one has to take into consideration is the overall diversity of the available hardware. The modern landscape of HPC and Datacenter Computing has drastically changed thanks to cloud server vendors (Amazon, Google, Microsoft, etc), researchers are now able to develop and run their applications on a wide variety of different CPUs and different GPUs. So the above question/dilemma can be stated as: "Given an OpenMP implementation of an application and the available CPUs and GPUs, is it worth proceeding to an OpenMP-to-CUDA transformation? What order of magnitude of performance improvement should one expect from such a transformation?" This diploma thesis aims to the development of a model that provides a performance and power prediction. More specifically, for an application our model predicts (a) the order of magnitude of the relative speedup that a specific GPU offers relatively to a specific CPU, (b) the most power efficient architecture between the two.	en
heal.advisorName	Γκούμας, Γεώργιος	el
heal.committeeMemberName	Γκούμας, Γεώργιος	el
heal.committeeMemberName	Κοζύρης, Νεκτάριος	el
heal.committeeMemberName	Πνευματικάτος, Διονύσιος	el
heal.academicPublisher	Εθνικό Μετσόβιο Πολυτεχνείο. Σχολή Ηλεκτρολόγων Μηχανικών και Μηχανικών Υπολογιστών. Τομέας Τεχνολογίας Πληροφορικής και Υπολογιστών	el
heal.academicPublisherID	ntua
heal.numberOfPages	114 σ.	el
heal.fullTextAvailability	false