yorik wrote:Wow, a cuda-based solver is interesting, must be blind fast!
Indeed, it is very interesting. CaeLinux 2013 OS already have implemented some patches from this website (to be more precise - you have to download them from one thread on caelinux.com forum). Exodus format make possible to postprocess in ParaView (results from CCX). In newer release they want to add also CUDA-Cusp.
...Crazy that GPUs can now be used for non-graphical tasks!
GPGPU is nothing very new, and now many developers are working hard to parallelize software for CPUs, GPUs and cards like Intel Xeon Phi. I have heard that software companies are switching from CUDA to OpenCL recently.
mk2 wrote:Gpgpu computing will be really interesting for simulation. But the meshing is still a problem in parallel
Depends on problem, some meshers already are working parallel and very fast, for example in ANSYS and many other apps (mostly commercial ones). One good opensource example is SnapyHexMesh (CFD mesher from OpenFOAM).
Take a look at AGROS2D (http://www.agros2d.org/
) - opensource Multiphysics (FEA/CFD/Acoustics/Magnetics,...) project. It doesn't use - so far - GPUs, but heavilly uses OpenMP and with very good results (it is pretty fast on modern CPUs).