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

Table 4

Full unrolling algorithm on a single GPU unit. 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
	1 GPU unit				CPU
	(16, 16, 1)	(128, 1, 1)	(256, 1, 1)	(512, 1, 1)
64^3	0.59/29.7	0.58/30.2	1.15/15.2	2.4/7.29	17.5
128^2×64	2.33/35.6	2.29/36.2	2.3/36.1	4.8/17.3	83
128^3	4.46/38.1	4.5/37.8	4.38/38.8	9.2/18.5	170
256^2×128	17.9/39	17.5/39.7	17.5/39.7	18.4/37.8	696
256^3	34.9/39.9	34.2/40.7	34.3/40.6	36/38.7	1393
512^2×128	71.4/39.4	70/40.2	70.1/40.1	73.9/38.1	2814
512^2×256	140/40.3	137/41.1	137/41	145/38.9	5623
512^3	276/40.9	271/41.7	271/41.7	286/39.5	11291