nanoporetech / dorado

Oxford Nanopore's Basecaller
https://nanoporetech.com/
Other
438 stars 53 forks source link

Dorado correct performance? #883

Open tnn111 opened 2 weeks ago

tnn111 commented 2 weeks ago

Are there any performance numbers for dorado correct? I'm interested in using it to correct ~120 Gbp (a full PromethION flow cell) and it'd be great to have some idea of the performance to expect. Like is it a 100 A100 hours or is it more like 1,000 :-) Thanks!

HalfPhoton commented 2 weeks ago

Dorado read correct performance is dominated by CPU usage during the alignment phase and the GPU is a relatively small part. Your A100 should be more than able to keep up with the GPU load. I would however recommend using a machine with the best CPU for this task even with a slightly weaker GPU.

We're actively looking at improving the performance of read correction and we will push updates as soon as we're able.

Best regards, Rich