Download
GPUSPH free surface flow model is written in C++ and CUDA and has been specifically developed on Linux and MacOS machines. To use GPUSPH you must have a CUDA-compatible NVIDIA graphics card (see www.nvidia.com/object/cuda_gpus.html to see if your graphics card runs CUDA). Your card must have at least Compute Capability of 1.1 to run GPUSPH. (Sorry, we don’t do Windows.)
NVIDIA’s CUDA Compiler and Driver
First, if you do not have CUDA installed on your computer, you must install the NVIDIA CUDA compiler from the NVIDIA website, Developer Zone. The version of CUDA Toolkit and CUDA Software Development Kit (SDK) for your operating system, found under Cuda Downloads, needs to be installed. (Note that the GPUSPH code offered below is CUDA 4.1 compatible.) These NVIDIA download packages include the CUDA compiler nvcc, which is needed to develop executable code, and the graphics card driver that allows your program to access the GPU card. Be sure to follow their instructions including setting DYLD_LIBRARY_PATH.
To ensure that all is installed correctly and working, you should compile and run the SDK examples, which include many NVIDIA programs that illustrate the various capabilities of CUDA and the GPU; for example, Nvidia's sorting program radixSort is used by GPUSPH to organize the neighbor list. To compile the SDK programs, after the SDK is installed, go to /Developer/GPU Computing/C and (on a unix/linux or mac machine), type make on a terminal window command line. This should create a directory of executable examples located within the C directory called bin/darwin/release for the mac and bin/linux/release for a linux machine. Some relevant SDK programs are fluidsGL and particles. In this directory, type ./fluidsGL to run the fluidsGL example. You should see a green window open on your desktop. Use the mouse to stir up the fluid. The example Particles, which has helped us write this code, is fun to play with as well.
GPUSPH