Skip to main content

Table 5 Acceleration performance of GPU on BWT construction

From: BASE: a practical de novo assembler for large genomes using long NGS reads

  Read_num Read_length With GPU Without GPU Time fold
S.aureus MW2 6,720,000 100 133 s 184 s 1.38
V.para 4,896,000 250 329 s 514 s 1.56
YH, 100 bp 1,057,750,382 100 5 h 19 h 3.80
NA12878, 150 bp 770,960,980 150 7 h 30 h 4.29
  1. To evaluate the acceleration performance of GPU on BWT construction, we used two versions, one of which used GPU and the other not, to build bi-directional BWT on four datasets. The results showed that especially in large genome dataset, compared with GPU version, version without GPU costs ~4 fold more time to construct BWT