Performance Optimization of 3D Lattice Boltzmann Flow Solver on a GPU

Table 3

Performance (MLUPS) comparisons of four implementations.

Domain sizes	TimeSteps	Serial	AoS	SoA_Pull_Only	SoA_Pull_Full_Tiling
	1000	9.89	111.73	759.52	1034
	5000	9.75	112.24	814.01	1115.63
	10000	9.82	111.73	818.36	1129.32
	Avg. Perf.	9.82	112.41	797.30	1092.99
	1000	7.64	78.58	798.21	1115.2
	5000	7.65	78.74	855.55	1189.15
	10000	6.98	78.74	861.42	1199.33
	Avg. Perf.	7.42	78.69	838.39	1167.69
	1000	9.47	76.79	811.35	1114.96
	5000	9.69	76.89	866.76	1185.95
	10000	9.56	76.91	871.39	1205.48
	Avg. Perf.	9.57	76.86	849.83	1168.8
	1000	8.99	74.74	787.76	1113.77
	5000	8.91	75.09	873.8	1182.33
	10000	8.9	775.14	883.56	1210.63
	Avg. Perf.	8.93	74.99	848.37	1168.91