======================================================
README.FIRST - BarraCUDA 0.7.0
======================================================
Thank you very much for downloading BarraCUDA. This software is a GPU port based on BWA version 0.6.1 and uses NVIDIA GPUs to accelerate the alignment of short sequencing reads from Next-generation Sequencers to a reference genome. This package contains the source code of the program and also a set of sample data from the NCBI Sequencing Read Archive (www.ncbi.nlm.nih.gov/sra) for testing purpose.
System Requirements
----------------------
Before installing BarraCUDA, please ensure that your computer has:
- A modern x86-based processor
- 8 GB of system memory (for large genomes)
- At least 20 GB of disk space (for large genomes and large sequencing libraries)
- An NVIDIA GPU with at least 1GB RAM (see below) and CUDA capability of 3.5 or above
- A Linux/UNIX operating system
VRAM Requirements
----------------------
BarraCUDA needs a memory space of at least the size of the BWT-encoded genome (the size of .bwt + .rbwt files) + another 675 MB of buffer to perform alignments.
The software also requires:
- NVIDIA CUDA toolkit SDK version 6 or above
- GCC and G++ Version 4.5 or above
- NVIDIA graphics driver version 340 or above
- zlib-devel (or zlib1g-dev) library
Note for indexing large genomes
-------------------------------
The default index command now selects the appropiate algorithms for indexing genomes. However you can still select
specific algorithm ('bwtsw' or 'is')for BWT indexing. The 'is' algorithm is quick but only work on small genomes,
for large genomes, like the human genome please use the bwtsw algorithm.
Support for older NVIDIA GPUs
-----------------------------
The current version (0.7.105) no longer supports legacy NVIDIA GPUS, please use older versions as follows:
CUDA Capability Version
2.0-3.0 BarraCUDA 0.6.2
1.3 BarraCUDA 0.5.1
<1.3 Not supported
Changes in 0.5 beta
----------------------
- initial release
Changes in 0.5.1 beta
----------------------
- Unified 32-bit and 64-bit source code
- Now works with CUDA toolkit version 3.2 and NVIDIA driver 260.19.21 even for full genomes (< 4 gigabases)
- Added support for NVIDIA Fermi cards
- Dropped support for older NVIDIA cards (now a GT200 based graphics with compute capability of 1.3 or above is required, e.g. GTX280, Tesla C1060 or above)
- Fixed an error in reporting alignment timings if compile with GCC 4.4.x
- Fixed the segmentation fault error when the program exits to the prompt
- Reduce maximum read length support from 125 to 80 bp to relief register pressure
- Added SAM headers in SAM output
Changes in 0.6 beta
----------------------
- Now works with CUDA toolkit 4.0 or above
- Magnitude of speed boost (12x) on long sequence reads compared to v0.5.1 beta, especially with Fermi cards
- Added memory detection
- Added seeding capability
- Default value for option '-o' changed to 1 (same as BWA), use '-o 0' to disable gap openings
- Increase max read length to 150 bp
Changes in 0.6.1 beta
----------------------
- NEW CUDA mode for samse core
- Enhanced memory detection
- Tested working with CUDA toolkit 4.1 or above
- Added automatic BWT size detection in 'index' core to automatically choose between 'is' or 'bwtsw' BWT building algorithm
- Added RG line support
- Improved infer size calculation for paired-end alignments
- Added BWA v0.5.x compatible .sai output option
Changes in 0.6.2 beta
----------------------
- New POSIX multithreading support for sampe core
- New fast alignment mode through aggressive ungapped seeding
- (R256) use '-f' to disable fast alignment mode
Changes in 0.7.105
----------------------
- Upgraded to BWA 0.6.x code base and uses the new bidirectional BWT index
- Updated CUDA alignment algorithm with much faster throughput
Changes in 0.7.107
----------------------
- Minor bug fixes
Compiling BarraCUDA
----------------------
To compile BarraCUDA, make sure you have GCC 4.5 and CUDA toolkit SDK version 6 (or above) installed in the system. The software should compile automatically by typing in the root directory of the source package:
make all
and the binary "barracuda" should be generated and placed in ‘linux/release’ folder.
To perform a quick test to see if BarraCUDA can work with the GPU in your system, you can type:
barracuda deviceQuery
and it should list out all the CUDA-capable devices available in your system, e.g. for a PC with a Tesla C1060,C2050 and a K40, it will list:
0 4095 13
1 3071 20
2 12287 35
The first number is the CUDA device number, followed by the amount of GPU memory in MB and the third is CUDA compute capability, 13 means 1.3, 20 means 2.0, 35 means 3.5.
Pre-compiled binaries
---------------------
Pre-compiled binaries are no longer available from this release
Sample data
-----------
Sample data is no longer available in the release, but can be downloaded from our SourceForge website
https://sourceforge.net/projects/seqbarracuda/files/Sample%20Data/
Acknowledgments
---------------
We want to thank Dr Bill Langdon, Dr Arran Schlosberg, Petr Klus, Simon Lam and Dag Lyberg who contributed to the code.
