On Feb 16, 2007, at 17:05 , Miguel Mart?nez Canales wrote: > I've been running some phonons calculations (Itanium 2, Linux, > probably MPICH) and was amazed to see huge differences between > serial espresso 3.1.1 runs and parallel 3.2.1 runs. I've narrowed the > search, though, and the problem seems to arise whenever I call ph.x > with the "-npool" suffix. Is it stupid to run ph.x the following way?
no, it is stupid that ph.x once again doesn't work with pools ... The problem is known (since yesterday), it should be limited to LSDA calculations, it is not yet fixed. There was also another problem with pools: if the number of pools is not a divisor of the number of k-points, pw.x may hang. This should be fixed in the current cvs. Paolo --- Paolo Giannozzi, Democritos and University of Udine, Italy
