Carlo Nervi wrote: > But I'm still not convinced that MPI is working 100% as it > should be (1500 s for the tests using a 8-core Xeon 5345).
tests are not made for benchmarking, even less for parallel execution: most of them are small and may take more time in parallel than in serial execution. As explained by Axel no less than 36 times, running parallel codes in multicore PCs can be tricky. > By the way, in one of my tests, using ./configure the > make.sys wrote FDFLAGS instead of FFLAGS. I was using > several switches, like LIBDIR, F90, F77, CC... and what should one get from such a vague statement? Paolo -- Paolo Giannozzi, Democritos and University of Udine, Italy
