Table 1. Benchmarking using real reads.

		GPU												CPU
Dataset	Volume (Gbp)	SOAP3-dp (Full SA)			SOAP3			BarraCUDA			CUSHAW			BWA			Bowtie2			SeqAlto			GEM			CUSHAW2
		%	Time (s)	Mem.	%	Time (fold)	Mem.	%	Time (fold)	Mem.	%	Time (fold)	Mem.	%	Time (fold)	Mem.	%	Time (fold)	Mem.	%	Time (fold)	Mem.	%	Time (fold)	Mem.	%	Time (fold)	Mem.
realYHPE100	12.24	98.12%	1,079	19	−10.60%	2.63	21.1	−6.39%	14.03	4.1	−3.13%	46.79	2.7	−4.14%	16.03	4.9	−3.87%	12.04	3.5	−0.48%	14.25	7.2	−2.86%	3.54	4.5	−1.39%	12.09	3.6
realYHPE150	56.23	97.16%	6,835	19.6	−28.33%	0.64	22.7	−10.99%	7.22	4.3	−10.11%	24.57	3.1	−8.05%	15.26	5.0	−7.45%	7.82	3.5	−3.65%	17.69	7.2	−5.85%	3.76	5	−6.47%	8.04	3.6
SRR211279	5.07	97.21%	439	18.4	−6.83%	1.05	20.8	−5.08%	13.42	4.1	−2.22%	54.22	2.1	−2.81%	16.27	4.9	−0.69%	12.00	3.5	−0.48%	17.29	7.2	−2.44%	4.11	4.6	−2.14%	12.03	3.6

The percentage of reads aligned and time consumption of aligners other than SOAP3-dp are recorded as the difference or ratio based on SOAP3-dp's figures. The ‘%’ column represents ‘Properly paired’ for PE reads. ‘Mem.’ represents the peak memory consumption in gigabytes during alignment.