I suggest you ...

Parallell simulations on GPU with RoadRunner (CUDA architecture)

It would be amazing to have support for parallelized simulations of RoadRunner on GPU, namely CUDA
http://docs.nvidia.com/cuda/cuda-c-programming-guide/#axzz3J8Gr6tok
I often have to solve the same SBML for different conditions/parameters (Monte-Carlo simulations). It would be nice if this could be run with RoadRunner on a GPU with CUDA architecture.

There is already some python interface to the cuda architecture which could be used (pyCUDA).
http://mathema.tician.de/software/pycuda/

For LSODA some basic implementation was done, but for CVODE nothing is available.
(Bioinformatics. 2011 Mar 15;27(6):874-6. doi: 10.1093/bioinformatics/btr015. Epub 2011 Jan 11.
GPU accelerated biochemical network simulation.
Zhou Y1, Liepe J, Sheng X, Stumpf MP, Barnes C.)

Especially for highly parallel problems this could be very interesting.

2 votes
Vote
Sign in
Check!
(thinking…)
Reset
or sign in with
  • facebook
  • google
    Password icon
    I agree to the terms of service
    Signed in as (Sign out)
    You have left! (?) (thinking…)
    Matthias KoenigMatthias Koenig shared this idea  ·   ·  Admin →

    1 comment

    Sign in
    Check!
    (thinking…)
    Reset
    or sign in with
    • facebook
    • google
      Password icon
      I agree to the terms of service
      Signed in as (Sign out)
      Submitting...
      • GerardGerard commented  · 

        OpenCL could be a better choice.
        The issue is more about the right choice of algorithm, as the speed boost is roughly inversely proportional to the algo complexity.
        I managed, with OpenCL, to run adaptative fifth order Runge-Kutta 200 time faster on a cheap HD 7970 Readon GPU than on a 2600k Intel CPU, but failed to find a stiff ODE solder adequate enough to be ported on GPU.
        As far as I see it, theoretical work on ODE solver focus on minimizing the number of model evaluations, while with GPU we could use large number of simultaneous evaluation at low cost.
        Give me a GPU compliant stiff ODE solver algo, I'll port it to OpenCL ;-)

      Feedback and Knowledge Base