I've tried doing alignment by cudamapper over 3 datasets, and compared to that by minimap2.
The time costs don't reduce much when compared to minimap2. In some cases, the running speeds of cudamapper are slower than minimap2, as shown in the below table.
Name
wall time(s)
mem peak(G)
note
data 1
cudamapper
650.63
10.87
v100, 16G
minimap2_v2.20
1687.86
31.53
-t 32 -k 17 -w 17 -x ava-ont
data 2
cudamapper
11193
39.76
v100, 16G
minimap2_v2.20
4491
19.21
-t 32 -k 17 -w 17 -x ava-ont
data 3
cudamapper
5958
27.4
NVIDIA TITAN xp, 12G
minimap2_v2.20
4590
54.02
-t 32 -x ava-ont
Also, the sensitivity of the cudamapper alignments is lower than that of minimap2, which leads to the poorer assembly results based on cudamapper alignments of the 3 above datasets. The algorithm of cudamapper might need to be modified to get alignment results similar to minimap2.
Hi,
I've tried doing alignment by cudamapper over 3 datasets, and compared to that by minimap2. The time costs don't reduce much when compared to minimap2. In some cases, the running speeds of cudamapper are slower than minimap2, as shown in the below table.
Also, the sensitivity of the cudamapper alignments is lower than that of minimap2, which leads to the poorer assembly results based on cudamapper alignments of the 3 above datasets. The algorithm of cudamapper might need to be modified to get alignment results similar to minimap2.