Parallel ODE solver

Can someone here tell me of any parallel CUDA implementation of a ODE solver capable of simulating Large System of coupled ODEs/DAEs.
The system arises from method of lines discretization of PDEs. Hope some of you experts here can give me some suggestions.

http://code.google.com/p/opencurrent/

http://code.google.com/p/opencurrent/

Actually this is a PDE solver on regular grid. However I need a ODE solver. But thanks for the information. My system of Coupled ODE has 60000 variables. So we need to find some type of parallel solver for such problem.

Actually this is a PDE solver on regular grid. However I need a ODE solver. But thanks for the information. My system of Coupled ODE has 60000 variables. So we need to find some type of parallel solver for such problem.

There probably isn’t a “general purpose” CUDA ODE solver you can use. All the necessary tools to build something for your specific problem exist (so sparse matrix-vector multiplication kernels, CG and BICGSTAB kernels for solving linear equations). Building an implicit or explicit sparse integrator shouldn’t be all that hard using something like CUSP as a starting point.

There probably isn’t a “general purpose” CUDA ODE solver you can use. All the necessary tools to build something for your specific problem exist (so sparse matrix-vector multiplication kernels, CG and BICGSTAB kernels for solving linear equations). Building an implicit or explicit sparse integrator shouldn’t be all that hard using something like CUSP as a starting point.

I also think that there is no “general purpose” ODE solver with CUDA.

But you can profile your ODE solver time to see if the linear algebra part takes a significant amount of time, if you, you can immediately take advantage of existing GPU accelerated linear algebra solvers, such as CULA and other free stuff.

I also think that there is no “general purpose” ODE solver with CUDA.

But you can profile your ODE solver time to see if the linear algebra part takes a significant amount of time, if you, you can immediately take advantage of existing GPU accelerated linear algebra solvers, such as CULA and other free stuff.