Table 3.

Time comparison between our solution and the Farrars' implementation

avg. length	# of sequences	CPU, Farrar	1 GPU	4 GPUs
51	4000	24,113	8,064	2,070
	8000	95,156	31,111	7,855
	12000	210,806	69,083	17,439
154	2000	28,300	17,931	4,609
	4000	112,109	67,747	17,284
	6000	251,730	149,030	37,622
257	2000	61,182	49,226	12,535
	4000	242,756	186,436	47,305
	6000	543,656	410,255	103,278
459	2000	149,269	155,631	39,478
	4000	594,976	594,140	149,539
	6000	1339,538	1332,831	333,593
608	800	41,675	50,222	12,840
	1200	92,776	106,840	27,406
	1600	164,135	191,793	48,463
1103	800	123,572	164,780	41,946
	1200	278,194	359,065	89,899
	1600	495,624	628,847	158,699

Mean times (in seconds) for the Smith-Waterman algorithm applied to different sets of sequences. Average lengths of sequences as well as cardinality of sets are given. The Farrar's implementation computes only scores while our GPU-based implementation computes scores and alignments.