Doc. Ing. Jiří Jaroš, Ph.D.

NANDAPALAN Nimalan, JAROŠ Jiří, TREEBY Bradley E. a RENDELL Alistair P. Implementation of 3D FFTs across Multiple GPUs in Shared Memory Environments. In: Proceedings of the Thirteenth International Conference on Parallel and Distributed Computing, Applications and Technologies. Beijing, 2012, s. 167-172. ISBN 978-0-7695-4879-1. Dostupné z: http://ieeexplore.ieee.org/document/6589258/
Jazyk publikace:angličtina
Název publikace:Implementation of 3D FFTs across Multiple GPUs in Shared Memory Environments
Strany:167-172
Sborník:Proceedings of the Thirteenth International Conference on Parallel and Distributed Computing, Applications and Technologies
Konference:The Thirteenth International Conference on Parallel and Distributed Computing, Applications and Technologies
Místo vydání:Beijing
Rok:2012
URL:http://ieeexplore.ieee.org/document/6589258/
ISBN:978-0-7695-4879-1
Soubory: 
+Typ Jméno Název Vel. Poslední změna
iconPDCAT2012.pdfPDCAT 2012 paper850 KB2013-01-04 04:15:29
^ Vybrat vše
S vybranými:
Klíčová slova
GPU; UVA; unified-virtual-address; multi-GPU; FFT; distributed; shared-memory;
Anotace
In this paper, a novel implementation of the distributed 3D Fast Fourier Transform (FFT) on a multi-GPU platform using CUDA is presented. The 3D FFT is the core of many simulation methods, thus its fast calculation is critical. The main bottleneck of the distributed 3D FFT is the global data exchange which must be performed. The latest version of CUDA introduces direct GPU-to-GPU transfers using a Unified Virtual Address space (UVA) that provides new possibilities for optimising the communication part of the FFT. Here, we propose different implementations of the distributed 3D FFT, investigate their behaviour, and compare their performance with the single GPU CUFFT and CPU-based FFTW libraries.
In particular, we demonstrate the advantage of direct GPU-to-GPU transfers over data exchanges via host main memory. Our preliminary results show that running the distributed 3D FFT with four GPUs can bring a 12% speedup over the single node (CUFFT) while also enabling the calculation of 3D FFTs of larger datasets. Replacing the global data exchange via shared memory with direct GPU-to-GPU transfers reduces the execution time by up to 49%. This clearly shows that direct GPU-to-GPU transfers are the key factor in obtaining good performance on multi-GPU systems.
BibTeX:
@INPROCEEDINGS{
   author = {Nimalan Nandapalan and Ji{\v{r}}{\'{i}} Jaro{\v{s}} and E.
	Bradley Treeby and P. Alistair Rendell},
   title = {Implementation of 3D FFTs across Multiple GPUs in Shared
	Memory Environments},
   pages = {167--172},
   booktitle = {Proceedings of the Thirteenth International Conference on
	Parallel and Distributed Computing, Applications and
	Technologies},
   year = {2012},
   location = {Beijing, },
   ISBN = {978-0-7695-4879-1},
   language = {english},
   url = {http://www.fit.vutbr.cz/research/view_pub.php.cs?id=10171}
}

Vaše IPv4 adresa: 54.163.209.109
Přepnout na IPv6 spojení

DNSSEC [dnssec]