3D Data Denoising via Nonlocal Means Filter by Using Parallel GPU Strategies

Table 5

Full unrolling algorithm on 2 GPU units. Execution times and speed-up values for several block size configurations and 3D datasets of normally distributed random numbers. Search and similarity windows have been set according to |𝑉_𝑖|=11³ and |_𝑗𝐵_𝑖|=3³.

Dataset size	Execution time/speed-up
	2 GPU units				CPU
	(16, 16, 1)	(128, 1, 1)	(256, 1, 1)	(512, 1, 1)
64^3	0.12/146	0.23/76.1	0.43/40.7	0.9/19.4	17.5
128^2×64	0.87/95.4	0.85/97.6	0.85/97.6	1.79/46.4	83
128^3	1.94/87.6	1.9/89.5	1.9/89.5	3.99/42.6	170
256^2×128	7.73/90	7.58/91.8	7.59/91.7	7.98/87.2	696
256^3	16.3/85.7	15.9/87.4	16/87.3	16.8/83	1393
512^2×128	30.9/91.1	30.3/92.8	30.3/92.7	32/87.9	2814
512^2×256	65/86.5	63.8/88.2	63.8/88.1	67.3/83.5	5623
512^3	133/84.8	131/86.4	131/86.3	138/81.8	11291