Measurement data for paper &quot;Evaluation of Asynchronous Offloading Capabilities of Accelerator Programming Models for Multiple Devices&quot;

Hahnfeld, Jonas; Price, James; Pflug, Hans Joachim; Müller, Matthias S.; Terboven, Christian

doi:10.18154/RWTH-2017-10493

Measurement data for paper "Evaluation of Asynchronous Offloading Capabilities of Accelerator Programming Models for Multiple Devices"

Hahnfeld, Jonas^RWTH* ; Terboven, Christian^RWTH* ; Price, James ; Pflug, Hans Joachim^RWTH* ; Müller, Matthias S.^RWTH*

2018

ImpressumAachen : RWTH Aachen

Online
DOI: 10.18154/RWTH-2017-10493
URL: https://publications.rwth-aachen.de/record/710311/files/data.zip

Einrichtungen

Projekte

User Support for Improved Scalability of Hybrid Parallel Codes (jara0001_20120501) (jara0001_20120501)

Kurzfassung
Accelerator devices are increasingly used to build large supercomputers and current installations usually include more than one accelerator per system node. To keep all devices busy, kernels have to be executed concurrently which can be achieved via asynchronous kernel launches. Our work compares the performance for an implementation of the Conjugate Gradient method with CUDA, OpenCL, and OpenACC on NVIDIA Pascal GPUs. Furthermore, it takes a look at Intel Xeon Phi coprocessors when programmed with OpenCL and OpenMP. In doing so, it tries to answer the question of whether the higher abstraction level of directive based models is inferior to lower level paradigms in terms of performance.This archive contains the modications to liboffload, all binaries and libraries including their respective commit ids, and the raw data of ourmeasurements.

OpenAccess:
ZIP

Dokumenttyp
Dataset

Format
online

Sprache
English

Interne Identnummern
RWTH-2017-10493
Datensatz-ID: 710311

Beteiligte Länder
Germany, UK

Related:

Contribution to a book/Contribution to a conference proceedings Hahnfeld, J. (Corresponding author)RWTH* ; Terboven, C. (Corresponding author)RWTH* ; Price, J. ; Pflug, H. J.RWTH* ; Müller, M. S.RWTH*
Evaluation of Asynchronous Offloading Capabilities of Accelerator Programming Models for Multiple Devices
2018Accelerator Programming Using Directives : 4th International Workshop, WACCPD 2017, Held in Conjunction with the International Conference for High Performance Computing, Networking, Storage and Analysis, SC 2017, Denver, CO, USA, November 13, 2017, Proceedings / edited by Sunita Chandrasekaran, Guido Juckeland
4. International Workshop on Accelerator Programming Using Directives, WACCPD 2017, Denver, CODenver, CO, USA, 13 Nov 2017 - 13 Nov 20172017-11-132017-11-13 Cham : Springer International Publishing, Lecture notes in computer science 10732, 160-182 (2018) [10.1007/978-3-319-74896-2_9]2018 BibTeX | EndNote: XML, Text | RIS