Skip to main content

Table 10 Running time comparison of CUDA and OpenCL implementations of MaxSSmap (denoted by MSS)

From: MaxSSmap: a GPU program for mapping divergent short reads to genomes with the maximum scoring subsequence

  (a) Time in minutes to map 100,000 251 bp reads to the E.coli genome. See Table 3 caption for details about reads
Div MSS_fast MSS MSS_fast MSS MSS_fast MSS
CUDA 4.2 CUDA 6.0   OpenCL  
Reads without gaps  
.1 20 28 17 27 17 27
.2 20 28 17 27 17 27
.3 20 28 17 27 17 27
Reads with gaps  
.1 20 28 17 27 17 27
.2 20 28 17 27 17 27
.3 20 28 17 27 17 27
   (b) Time in minutes to map paired human reads from NA12878 in 1000 genomes (SRR016607) of length 101 bp to the human
   genome. We denote NextGenMap by NGM and MaxSSmap by MSS. See Table 8 for more details about reads
NGM+ NGM+ NGM+ NGM+ NGM+ NGM+  
MSS_fast MSSmap MSS_fast MSS MSS_fast MSS  
CUDA 4.2   s   OpenCL   
1295.9 2242.4 1183.5 2092.7 1252.7 2159.9  
   (c) Time in minutes to map 100,000 ancient horse DNA reads (SRR111892) of length 76 bp to
   the horse genome Equus_caballus EquCab2 (GCA_000002305.1). See Table 9 for more details
   about reads
NGM+ NGM+ NGM+ NGM+ NGM+ NGM+  
MSS_fast MSSmap MSS_fast MSS MSS_fast MSS  
CUDA 4.2   CUDA 6.0   OpenCL   
1609.6 2836 1515 2689.8 1561.8 2736.8  
  1. The output from the three methods give the same accuracies and errors as given earlier but the running times vary. We find the CUDA 6.0 implementation to have the lowest runtimes followed by OpenCL and CUDA 4.2.