DocumentCode :
2190114
Title :
Graphics Card Computing for Cosmology: Cholesky Factorization
Author :
Gratton, Steven
Author_Institution :
Kavli Inst. for Cosmol., Univ. of Cambridge, Cambridge, UK
fYear :
2010
fDate :
June 29 2010-July 1 2010
Firstpage :
1207
Lastpage :
1212
Abstract :
Cosmological data sets are becoming so large as to make optimal statistical analyses of them impossible. Even with approximations made, the computational challenges can be severe. Cholesky Factorization of matrices is an essential tool. This paper reports on progress made by the author in implementing Cholesky Factorization on one or more graphics processing units (GPUs). Particular attention is paid to the process of the development of the algorithm in light of the characteristic features of the GPUs. Performance and optimization are discussed, and potential avenues for future development are presented.
Keywords :
astronomy computing; computer graphic equipment; coprocessors; cosmology; matrix decomposition; Cholesky matrix factorization; GPU; cosmological data sets; cosmology; graphic card computing; graphic processing units; optimal statistical analyses; Bandwidth; Data models; Graphics processing unit; Instruction sets; Kernel; Strips; Tiles; Determinants; Matrix Inversion; Parallel algorithms; Parallel and vector implementation;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Computer and Information Technology (CIT), 2010 IEEE 10th International Conference on
Conference_Location :
Bradford
Print_ISBN :
978-1-4244-7547-6
Type :
conf
DOI :
10.1109/CIT.2010.218
Filename :
5577889
Link To Document :
بازگشت