On Nov 30, 2010, at 5:32 PM, Synge Todo wrote:
On 2010/12/01, at 2:15, degliesposti wrote:
I wonder if some of you is experienced and could give us recent insights about these two points:
i) core parallelization with ALPS: which parallel environment is better (OpenMP, MPICH, ...)
It depends on the machines and the algorithms you use.
The ALPS (conventional) scheduler supports only for parameter parallelization using MPI. On the other hand, the ALPS/parapack scheduler also supports MPI, OpenMP, and hybrid (MPI + OpenMP) parallelization inside the workers. Please look at the examples in example/parapack/single and example/parapack/multiple directories to learn how to write parallel workers using MPI or OpenMP.
ii) are there any porting/tests of ALPS on GPU's ?
I don't know if there are any activities about ALPS on GPGPU.
ALPS 2.1, planned for a year from now, should include some optimizations for GPUs in DMRG codes and maybe others.
Matthias