NGS with GPUs

[SMcTaggart]

Hello;

We are in the process of deciding on our compute hardware for processing data from a MiSeq and a MinION. There was a question posted in 2011 about the utility of using GPUs versus CPUs. I am wondering how many GPU compatible bioinformatic software packages are now available. Is anyone using an exclusively GPU system for their NGS processing requirements?

Cheers!

Seanna

[Brian Bushnell]

Nvidia has made some efforts to port a few bioinformatics algorithms to GPUs. But that vast majority of bioinformatics software runs only on CPUs, so planning on a GPU-oriented hardware purchase would greatly cripple your flexibility. Our compute cluster has a few thousand CPU cores, and zero GPUs; and nobody really seems to miss their absence.

[blancha]

I process NGS data all day long.
I have never used the GPU cores available on our cluster.
I'd be really curious to know though what the potential benefits of using software capable of running on the the GPU could be though.

[colindaven]

Certainly for MiSeq + Minion data you won't need GPUs to scale up. There are a number of GPU programs out there

eg

https://www.cebitec.uni-bielefeld.de/brf/index.php/software/27-saruman-homepage

https://github.com/masayano/CLAST

https://github.com/aquaskyline/SOAP3-dp

However, most are not end-to-end solutions or have other drawbacks.

I would stick to CPU as I do for 99.9% of my NGS work.

[cement_head]

I process NGS data all day long.
I have never used the GPU cores available on our cluster.
I'd be really curious to know though what the potential benefits of using software capable of running on the the GPU could be though.

I'm very interested in GPU-BLAST, which according to the OSC (Ohio Supercomputing Center) is at least four fold faster than CPU based BLASTing. That alone would help speed up our transcriptome annotations. The new batch of software, along with NVBIO seems to be promising: https://www.nvidia.com/en-us/data-ce...tions/catalog/