GPU-BLAST

GPU-BLAST is an accelerated version of the popular NCBI-BLAST (www.ncbi.nlm.nih.gov) that uses general-purpose graphics processing unit (GPU). In comparison to the sequential NCBI-BLAST, GPU-BLAST is nearly four times faster, while producing identical results.

Availability & Restrictions

GPU-BLAST is available without restriction to all OSC users.

The following versions of GPU-BLAST are available on OSC systems:

VERSION NCBI-BLAST VERSION OAKLEY RUBY
1.0 2.2.24+ X  
1.1 2.2.25+ X  
1.1 2.2.26+  X*  
1.1 2.2.28+   X*

 

* Current default version

Note that GPU-BLAST is not available on the Owens system. GPU-BLAST requires the CUDA nvcc compiler, but the GPU-BLAST software is only compatible with CUDA versions up to 7.5. CUDA 8, installed on the Owens cluster, is incompatible with GPU-BLAST.

Usage

Usage on Oakley

Set-up

To set up the environment for gpu-blast on Ruby cluster, do the following:

module load cuda/4.1.28
module load gpu-blast

Batch Usage

Provided is a sample batch script:

#PBS -l nodes=1:ppn=1:gpus=1
#PBS -l walltime=10:00
#PBS -N GPU-Blast
#PBS -S /bin/bash
#PBS -j oe
module load gpu-blast
module load cuda/4.1.28
set -x
cd $PBS_O_WORKDIR
mkdir $PBS_JOBID
cp 100.fasta $TMPDIR
cd $TMPDIR
/usr/bin/time blastn -db nt -query 100.fasta -out test.out
cp * $PBS_O_WORKDIR/$PBS_JOBID
 

Usage on Ruby

Set-up

To set up the environment for gpu-blast on Ruby cluster, do the following:

module load cuda/7.5.18
module load gpu-blast

Batch Usage

Provided is a sample batch script:

#PBS -l nodes=1:ppn=1:gpus=1
#PBS -l walltime=10:00
#PBS -N GPU-Blast
#PBS -S /bin/bash
#PBS -j oe
module load gpu-blast
module load cuda/7.5.18
set -x
cd $PBS_O_WORKDIR
mkdir $PBS_JOBID
cp 100.fasta $TMPDIR
cd $TMPDIR
/usr/bin/time blastn -db nt -query 100.fasta -out test.out
cp * $PBS_O_WORKDIR/$PBS_JOBID
 

Further Reading

See Also

Service: 
Fields of Science: