Heterogene Rechnerarchitekturen Online (HETRON)5 ECTS
(englische Bezeichnung: Heterogeneous Computing Architectures Online)
(Prüfungsordnungsmodul: Heterogene Rechnerarchitekturen Online)

Lehrveranstaltungen:

• • Heterogene Rechnerarchitekturen Online
    (Vorlesung, Dietmar Fey, Dies ist eine Online-Vorlesung: https://www.studon.fau.de/crs4119544_join.html)

Inhalt:

Whereas heterogeneous architectures and parallel computing has filled an academic niche in the past it has become now a commodity technique with the rising of multi-core processors and programmable graphic cards. Even FPGAs play a role hereby in a certain extent due to their increasing importance as accelerator hardware what is clearly observable in the scientific community. However, on one side parallel hardware like multi-core and GPUs are now available nearly for everybody and not only for a selected selection of people, who have access to a parallel supercomputer. On the other side the knowledge about programming of this commodity hardware, and we mean here in particular hardware-orientated programming in order to squeeze out all offered GFlops and TFlops of such hardware, is still missing as well as the knowledge about the architecture details. To overcome this lack we offer this course HETRON.
The e-learning course HETRON for the exploitation of parallel and heterogeneous computer architectures) focuses on two main topics which are closely related to each other. This concerns on one side the benefits of using different kinds of multi-core processors and parallel architectures built-up on base of these multicore processors. These architectures differ among each other in the number and in the complexity of its single processing nodes. We distinguish between systems consisting of a large number of simpler, so called fine-grained, processor cores vs. systems consisting of a smaller number of more complex, so called coarse-grained, processor cores. On the other side we lay our focus on that we want to do with these different heterogeneous parallel architectures, namely the execution of parallel programs. Of course this requires the use of parallel programming languages and environments, like CUDA or OpenMP. However, besides these questions of using the right syntax and the right compiler switches to optimize a parallel program it is a pre-requisite to understand how parallel computing really works. This refers (i) to the comprehension which basic mechanisms of parallel computing exist, (ii) where are the limits of getting more performance with parallel computing and (iii) in what context stand these mechanisms to heterogeneous architectures. In other words it handles the question which architecture is the best one for a certain parallelization technique. To teach these three topics, is one main goal we pursuit with the course HETRON, and of course, this more fundamental basics of heterogeneous and parallel computing have to be proven by means of concrete application examples to deepen the acquired knowledge about heterogeneous architectures and parallel computing principles.

Lernziele und Kompetenzen:

Die Studierenden ...
...verstehen die Notwendigkeit sowie grundlegende Anwendungsfälle für heterogene Rechnerarchitekturen.
...können den grundlegenden Aufbau und das Zusammenspiel der Komponenten heterogener Rechnerarchitekturen erklären.
...erläutern grundsätzliche Parallelisierungsprinzipien wie Amdahls Law, High-Performance- und High-Througput-Computing sowie Parallelisierungsstrategien.
...können einfache Programme mit Hilfe der vermittelten Prallelisierungsprinzipien (Amdahls Law, High-Performance- und High-Throughput-Computing) analysieren und entsprechende Parallelisierungsstrategien entwickeln.
...erklären den Aufbau sowie Stärken und Schwächen von verschiedenen Architekturen wie CPUs, GPUs, Many-Core Prozessoren und FPGAs.
...implementieren ausgewählte Anwendungsbeispiele (SHA256 Algorithmus, Ising-Modell und Fast-Fourier-Transformation) auf oben genannte Architekturen.
...erforschen und bewerten verschiedener Parallelsierungstechniken in Abhängigkeit der Anwendung und der Architektur.
...erläutern die Grundlagen des Grid- und Cloud-Computings
...sind in der Lage parallele Berechnungen (SHA256) im Grid umzusetzen.

Verwendbarkeit des Moduls / Einpassung in den Musterstudienplan:

1. Computational Engineering (Rechnergestütztes Ingenieurwesen) (Master of Science)
   (Po-Vers. 2013 | TechFak | Computational Engineering (Rechnergestütztes Ingenieurwesen) (Master of Science) | Gesamtkonto | Wahlpflichtbereich Informatik | Wahlpflichtbereich Informatik | Heterogene Rechnerarchitekturen Online)

Studien-/Prüfungsleistungen:

Heterogene Rechnerarchitekturen Online (Prüfungsnummer: 275245)

(englischer Titel: Heterogeneous Computing Architectures Online)

Prüfungsleistung, mündliche Prüfung, Dauer (in Minuten): 30, benotet, 5 ECTS

Anteil an der Berechnung der Modulnote: 100.0 %

weitere Erläuterungen:
Die Prüfungssprache ist abhängig von der Wahl der Studierenden.

Prüfungssprache: Deutsch oder Englisch

Erstablegung: WS 2022/2023, 1. Wdh.: SS 2023

1. Prüfer:

Dietmar Fey