|Subject:||Re: [ESPResSo-users] electrical field+ lattice boltzmann|
|Date:||Sat, 27 Dec 2014 22:09:04 +0100|
To my knowledge, only charged species move along an electric field. Which net charge do you think your fluid molecules carry?
What do you mean with "parallel on gpu”? Even the smallest available CUDA devices have at least 32 compute cores, which we need to feed, so LBPUG always runs “parallel on gpu”.
Now, if you mean to run parallel on several GPUs, then this isn’t possible, as ESPResSo doesn’t have multi-GPU support yet.
However, you can still use MPI to run the CPU part in parallel. In this case, ESPResSo assumes that the GPU is connected to rank 0, and all non-GPU algorithms during propagation parallelize over the available MPI cores. For running the CPU parts in parallel, you first need to compile with an MPI compiler. Usually this is mpicxx, but that highly depends on your system, in particular if you are running on bigger clusters/supercomputers. After that you typically run “mpiexec -n <cores> Espresso <script>” to parallelize over <cores> cores, but again, that depends on your MPI implementation or queueing system.
During configure, Espresso is checking for some common names of the MPI compiler, so if it says at the end of running configure that MPI is disabled, then you need to figure out how to compile against MPI on your specific system. On standard Linux systems, the MPI is usually openmpi, but it may just not be installed. On larger clusters, you usually find in the documentation pages how to compile against MPI. Often, you need to load certain modules or similar. Here, we can’t help you, that is up to the system’s administrators.
JP Dr. Axel Arnold
ICP, Universität Stuttgart
70569 Stuttgart, Germany
Phone: +49 711 685 67609
|[Prev in Thread]||Current Thread||[Next in Thread]|