[Beowulf] GPU-HMMer for interested people

[Olli-Pekka Lehto]

Joe Landman wrote:
> Hi folks
> 
>   GPU-HMMer (part of the MPI-HMMer effort) has just been
> announced/released at http://www.mpihmmer.org
> 
>   MPI-HMMer has itself been improved with parallel-IO and better
> scalability features.  JP has measured some large number (about 180x)
> over single cores on a cluster for the MPI run.
> 
>   Enjoy!
> 
> Joe
> 

Hi Joe,

Looks quite promising. Here are results from a simple real-world test case:

GPU: Dual GTX280, each with 1GB RAM
CPU: Single Intel Core2 quad Q9550 2.83GHz

hmmsearch 4 threads sorted:        		274.49s
hmmsearch 4 threads unsorted:     		254.23s
cuda_hmmsearch unsorted            		407.85s
cuda_hmmsearch sorted:              		62.69s
cuda_hmmsearch sorted 2 simultaneous runs:	78.23s 80.79s

Remarks:

-Running hmmsort to sort the sequence database is critical to obtain 
reasonable performance from cuda_hmmsearch. However, the regular 
hmmsearch is slightly slower with the sorted database.

-Running two simultaneous runs assigned to different GPUs on a dual-GPU 
quad-core system yields some performance penalty, but is still quite 
feasible.

-I used the parameters THREADSIZE=320 BLOCKSIZE=64. I'm not completely 
sure if these are the optimum values for GTX280. Any better suggestions?

Regards,
Olli-Pekka
-- 
Olli-Pekka Lehto, Systems Specialist, Special computing, CSC
PO Box 405 02101 Espoo, Finland; tel +358 9 4572215, fax +358 9 4572302
CSC is the Finnish IT Center for Science, www.csc.fi,
e-mail: Olli-Pekka.Lehto at csc.fi