This paper presents a fast and robust approach to evaluate the singular values of a very large number of small matrices on GPUs. The timings and the accuracy obtained are compared to those achieved by "state-of-the-art" libraries. The results show that, by suitably tuning the trade-off between computational speed and accuracy, a reasonable speedup over a fast multi-core CPU is possible.
Massive computation of singular values of small matrices on GPUs / Capozzoli, Amedeo; Curcio, Claudio; Liseno, Angelo; DI DONNA, Salvatore. - (2015), pp. 36-37. (Intervento presentato al convegno Computational Electromagnetics International Workshop, CEM 2015 tenutosi a Izmir, Turkey nel July 1-4, 2015) [10.1109/CEM.2015.7237428].
Massive computation of singular values of small matrices on GPUs
CAPOZZOLI, AMEDEO;CURCIO, CLAUDIO;LISENO, ANGELO;DI DONNA, SALVATORE
2015
Abstract
This paper presents a fast and robust approach to evaluate the singular values of a very large number of small matrices on GPUs. The timings and the accuracy obtained are compared to those achieved by "state-of-the-art" libraries. The results show that, by suitably tuning the trade-off between computational speed and accuracy, a reasonable speedup over a fast multi-core CPU is possible.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
