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

Table 3

Partial 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	0.49/35.7	0.38/46.1	0.66/26.5	1.36/12.9	17.5
128^2×64	1.15/72.2	1.12/74.1	1.11/74.8	2.27/36.6	83
128^3	2.31/73.6	2.25/75.6	2.23/76.2	4.57/37.2	170
256^2×128	8.44/82.5	8.12/85.7	8.28/84.1	9.26/75.2	696
256^3	16.9/82.5	16.2/85.8	16.6/84.1	18.6/74.7	1393
512^2×128	32.6/86.2	31.7/88.9	31.92/88.2	34.5/81.3	2814
512^2×256	65.2/86.2	63.6/88.5	63.9/88	69.2/81.3	5623
512^3	131/86.5	129/87.8	128/87.9	139/81.2	11291