[marcowanger]

We have purchased Novoalign and buying GTX690 possibily. Hopefully we can test if SOAP3 really outperforms BWA and Novoalign

Novoalign is pretty quick but it costs money and is not open-source so it is understandable that not everyone wants to use it.

Anyways, I think most of us agree in general on this debate, however we just don't have enough information about speed and accuracy to decide if it is worth using SOAP3. Hopefully this will change soon

[adaptivegenome]

We have purchased Novoalign and buying GTX690 possibily. Hopefully we can test if SOAP3 really outperforms BWA and Novoalign

Post results! I do use BWA and Novoalign but I have not tried SOAP3. Did you try any of the BWA-based GPU mappers? Like Barracuda? That might also be a good comparison for SOAP3... both on speed and accuracy.

[colindaven]

According to what I have seen so far many GPU mappers seem to have serious problems, i.e. not producing SAM output. Also, this rather sceptical but also very insightful piece on the CPU/GPU debate provides some food for thought, whichever side of the fence you're on.

https://larsjuhljensen.wordpress.com/2011/01/28/commentary-the-gpu-computing-fallacy/

[marcowanger]

Post results! I do use BWA and Novoalign but I have not tried SOAP3. Did you try any of the BWA-based GPU mappers? Like Barracuda? That might also be a good comparison for SOAP3... both on speed and accuracy.

Will do when I have access the the licence and the GPU with new server

Have been BWAing for 6 weeks for a project...I also hope to do some GPU mapping ....

[ymc]

We have purchased Novoalign and buying GTX690 possibily. Hopefully we can test if SOAP3 really outperforms BWA and Novoalign

Based on my understanding of SOAP3's documentation, you need at least 3GB per GPU to run SOAP3. The only GeForce cards can do that now are GTX580 3GB and GTX680 4GB.

[xied75]

According to what I have seen so far many GPU mappers seem to have serious problems, i.e. not producing SAM output. Also, this rather sceptical but also very insightful piece on the CPU/GPU debate provides some food for thought, whichever side of the fence you're on.

https://larsjuhljensen.wordpress.com...uting-fallacy/

Thanks for this link, that's exactly what I was trying to say.

Here some links from me:

http://hothardware.com/News/Nvidia-Rails-Against-Intels-ManyCore-Architecture-/

http://www.dannzfay.com/2012/04/intels-knights-corner-mic-architecture.html

Another link a bit deep into the programming side:

http://www.yosefk.com/blog/simd-simt-smt-parallelism-in-nvidia-gpus.html

[lh3]

Based on my understanding of SOAP3's documentation, you need at least 3GB per GPU to run SOAP3. The only GeForce cards can do that now are GTX580 3GB and GTX680 4GB.

Maybe they want to put the FM-index on the graphical card? (don't know) BTW, GTX690 does have 4GB memory:

http://videocardz.com/nvidia/geforce...eforce-gtx-690

[lh3]

Thanks for this link, that's exactly what I was trying to say.

Here some links from me:

http://hothardware.com/News/Nvidia-Rails-Against-Intels-ManyCore-Architecture-/

http://www.dannzfay.com/2012/04/inte...hitecture.html

Thanks. I wonder if those 50 cores are symmetric. If they are non-symmetric, most existing algorithms will not have the full power. Honestly, I am not so interested in GFLOPS. GPU certainly has much better GFLOPS, but if we cannot achieve that performance for our tasks, talking about GFLOPS does not help much.

[nilshomer]

Maybe they want to put the FM-index on the graphical card? (don't know) BTW, GTX690 does have 4GB memory:

http://videocardz.com/nvidia/geforce...eforce-gtx-690

Random memory access is expensive on the GPU, and so algorithms have to be modified to limit the access and reduce any heaps/stacks. For example, the GPU version of BWA uses DFS to simulate the BFS in the short-read algorithm.

[xied75]

On wikipedia the description on code name Knights Ferry "In the Aubrey Isle core a 1,024-bit ring bus (512-bit bi-directional) connects processors to main memory." Thus this is a SMP system.

On Intel http://software.intel.com/en-us/blog...cture-support/ it says "SMP on a chip".

And here is the software stack. http://software.intel.com/en-us/blog...oftware-stack/

[ymc]

Maybe they want to put the FM-index on the graphical card? (don't know) BTW, GTX690 does have 4GB memory:

http://videocardz.com/nvidia/geforce...eforce-gtx-690

But it has two GK104 GPUs and it is 2GB per GPU. From CUDA's point of view, they are two devices with 2GB each. They have two different address spaces.

[sparks]

Intil Xeon Phi

Thanks. I wonder if those 50 cores are symmetric. If they are non-symmetric, most existing algorithms will not have the full power. Honestly, I am not so interested in GFLOPS. GPU certainly has much better GFLOPS, but if we cannot achieve that performance for our tasks, talking about GFLOPS does not help much.

I think there's some hyperthreading so it will be interesting to see.The coprocessor runs Linux and has MPI so hopefully all we need to change in Novoalign is the SSE routines so it should be an easy port to the Phi. we'll be trying this as soon as possible.

[Aqua]

Formal release for SOAP3-dp

SOAP3-dp is now formally published in PLoS ONE http://www.plosone.org/article/info%...l.pone.0065632. And the software available at http://i.cs.hku.hk/2bwt-tools/soap3-dp/ is with major bugs fixed and work well with downstream analysis tools.