search

isovic / racon

Ultrafast consensus module for raw de novo genome assembly of long uncorrected reads. http://genome.cshlp.org/content/early/2017/01/18/gr.214270.116 Note: This was the original repository which will no longer be officially maintained. Please use the new official repository here:
https://github.com/lbcb-sci/racon
MIT License
268 stars 48 forks source link

cuda usage #133

Closed michieitel closed 5 years ago

michieitel commented 5 years ago

Hi!

Any recommendations on how to use the cuda options?

-c, --cudapoa-batches default: 1 number of batches for CUDA accelerated polishing -b, --cuda-banded-alignment use banding approximation for polishing on GPU. Only applicable when -c is used. --cudaaligner-batches (experimental) Number of batches for CUDA accelerated alignment

How can I identify suitable settings for my GPU?

thanks Michael

rvaser commented 5 years ago

Hi Michael, I think that each batch is around 200MB, the more batches the faster it will run. Just divide your GPU memory in MB with 200 and round down. For further speed-up, you can use the option -b which might have a tiny bit of accuracy loss.

Best regards, Robert

michieitel commented 5 years ago

I see! Thanks for this information!

Any guess when you manage to implement short read polishing with cuda? That would be amazing!!

Best Michael

rvaser commented 5 years ago

You can ask the peeps from Nvidia @ https://github.com/clara-genomics/ClaraGenomicsAnalysis :)

Best regards, Robert

michieitel commented 5 years ago

Ah I see ;-)