[Pw_forum] PW taskgroups and a large run on a BG/P

[Paolo Giannozzi]

David Farrell wrote:

> 432 atoms (960 electrons) on 1024, 4096, 8192 procs
> 576 atoms (1280 electrons) on 1024, 2048, 4096 procs
> 864 atoms (1920 electrons) on 1024, 2048, 4096 procs

I recently run a test (with CP) of 838 atoms and 2311 electrons
on as little as 64 processors (and I think it would also fit into
32 processors). 8192 processors for 432 atoms is way off mark,
unless you can use other parallelization levels.

> - ndiag > 1 doesn't work in any case is pointing toward the idea that 
> the parallel orthogonalization isn't working. 

last time I tried on a BG, it was working (after I fixed a problem,
which was on the BG side, not on Q-E side). Please provide a test
job (the smaller you have, of course)

Paolo
-- 
Paolo Giannozzi, Democritos and University of Udine, Italy