Table 4.

Computation time t in μs and parallel efficiency E of the Fourier transform of 10 2D vectors of side length N calculated using the cuFFT batched mode.

N	1 GPU	2 GPUs		3 GPUs		4 GPUs
	t	t	E	t	E	t	E
384	429	230	0.93	189	0.76	152	0.71
432	532	280	0.95	228	0.78	178	0.75
486	730	387	0.94	320	0.76	251	0.73
512	555	288	0.96	233	0.79	179	0.78
784	1699	865	0.98	697	0.81	529	0.80